Chaos, Solitons and Fractals 140 (2020) 110103

Contents lists available at ScienceDirect

Chaos, Solitons and Fractals

Nonlinear Science, and Nonequilibrium and Complex Phenomena
journal homepage: www.elsevier.com/locate/chaos

Global stability and cost-effectiveness analysis of COVID-19

considering the impact of the environment: using data from Ghana
Joshua Kiddy K. Asamoah a,b,∗, Mark A. Owusu b, Zhen Jin a, F. T. Oduro b,c, Afeez Abidemi d,
Esther Opoku Gyasi b
a
Complex Systems Research Center, Shanxi University, Taiyuan 030006, China
b
African Institute for Mathematical Sciences, Accra-Ghana
c
Kwame Nkrumah University of Science and Technology, Kumasi-Ghana
d
Department of Mathematical Sciences, Universiti Teknologi Malaysia, Johor, 81310 Johor Bahru, Malaysia

a r t i c l e i n f o a b s t r a c t

Article history: COVID-19 potentially threatens the lives and livelihood of people all over the world. The disease is
Received 19 June 2020 presently a major health concern in Ghana and the rest of the world. Although, human to human trans-
Revised 6 July 2020
mission dynamics has been established, not much research is done on the dynamics of the virus in the
Accepted 9 July 2020
environment and the role human play by releasing the virus into the environment. Therefore, investi-
Available online 10 July 2020
gating the human-environment-human by use of mathematical analysis and optimal control theory is
Keywords: relatively necessary. The dynamics of COVID-19 for this study is segregated into compartments as: Sus-
COVID-19 ceptible (S), Exposed (E), Asymptomatic (A), Symptomatic (I), Recovered (R) and the Virus in the environ-
Environmental transmission ment/surfaces (V). The basic reproduction number R0 without controls is computed. The application of
Global stability Lyapunov’s function is used to analyse the global stability of the proposed model. We fit the model to real
Sensitivity analysis data from Ghana in the time window 12th March 2020 to 7th May 2020, with the aid of python program-
Optimal control
ming language using the least-squares method. The average basic reproduction number without controls,
Cost-effectiveness analysis
Ra0 , is approximately 2.68. An optimal control is formulated based on the sensitivity analysis. Numerical
simulation of the model is also done to verify the analytic results. The admissible control set such as:
effective testing and quarantine when boarders are opened, the usage of masks and face shields through
media education, cleaning of surfaces with home-based detergents, practising proper cough etiquette and
fumigating commercial areas; health centers is simulated in MATLAB. We used forward-backward sweep
Runge-Kutta scheme which gave interesting results in the main text, for example, the cost-effectiveness
analysis shows that, Strategy 4 (safety measures adopted by the asymptomatic and symptomatic individ-
uals such as practicing proper coughing etiquette by maintaining a distance, covering coughs and sneezes
with disposable tissues or clothing and washing of hands after coughing or sneezing) is the most cost-
effective strategy among all the six control intervention strategies under consideration.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction while China’s confirmed cases were exponentially increasing sum-

ming the total confirmed cases to 1320 as of 25th January [3]. Ger-
The COVID-19 is a communicable disease which is caused by many joined the race on 28th January with reported case of 1 [4].
coronavirus of a new kind. As of 21st January 2020, only 4 coun- Italy also reported its first 2 confirmed cases on 31st January 2020
tries namely: China (278 cases), Republic of Korea (1 case), Thai- contributing to 9826 globally confirmed cases [5]. The first-time a
land (2 cases) and Japan (1 case) were affected with a total of 282 Health Care Worker was diagnosed with coronavirus happened in
confirmed cases including 6 deaths [1]. The United States recorded France on 1st February 2020. Spain and the United Kingdom re-
its first case on 23rd January 2020 as a result of an exported ported their first confirmed cases on the same day each with 1
case from China [2]. This amounted to a total of 581 confirmed and 2 cases respectively [6]. Philippines recorded the first death
cases. France and Australia also reported 3 confirmed cases each outside China on 1st February 2020 [7]. The first African coun-
try to report a confirmed case was Egypt on 14th February 2020
∗
[8]. The statistics as of 19th June, 2020, 06:25 GMT showed that,
Corresponding author.
there were 8,586,115 confirmed cases, 456,458 confirmed deaths
E-mail addresses: [email protected] (J.K.K. Asamoah), [email protected]
(M.A. Owusu), [email protected] (Z. Jin), [email protected] (F. T. Oduro), and 4,535,245, recovered cases globally [9]. Ghana recorded its first
[email protected] (A. Abidemi), [email protected] (E.O. Gyasi). confirmed cases on 12th March, 2020 [10]. These were 2 imported

https://doi.org/10.1016/j.chaos.2020.110103
0960-0779/© 2020 Elsevier Ltd. All rights reserved.
2 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Fig. 1. Daily Infected trend and bar plot for confirmed cases for Ghana.

cases from Norway and Turkey. Prior to the identification of these ticed that, 6% of the secondary outcomes can be linked to environ-
cases, effective measures such as testing and public education had mental transmission. They also noted that, presymptomatic trans-
been put in place to create awareness which demonstrated the missions can sufficiently maintain the epidemic growth. Hence, we
country’s preparedness to face this pandemic [11]. Earlier measures want to study the mathematical dynamics of viral load in the en-
after the first incidence included; a ban on individuals from coun- vironment and the use of non-pharmaceutical measures together
tries which had recorded over 200 cases and 14 days mandatory with cost-effectiveness analysis. Thus, by developing a mathemat-
quarantine for all persons who were allowed to enter its territory. ical and optimal control model for human-environment-human
Also, as part of the early measures against the spread, schools, con- transmission dynamics. Despite that other optimal control model
ferences, social gatherings, sporting activities and political rallies on COVID-19 have been studied (see for example [27–32]). But, to
were all halted as enhanced measures against the spread of COVID- the best of our knowledge, there exists no compartmental model
19 [12]. On 30th March 2020, Accra and Kumasi which are the ma- which focuses on environment to human dynamics of COVID-19 in
jor cities in the country experienced lockdown together with some Ghana together with cost-effectiveness analysis. Hence, this paper
internal travel restrictions. Ghana had recorded 12,929, confirmed will contribute to the existing knowledge on the spread of COVID-
cases including 66 deaths and 4,468 recovered with 8,395 active 19 and the dynamical influence of the virus in the environment
individuals as of 18th June, 2020 [11]. It is important to note that through human activities in Ghana. It will also assist health care
these deaths were also attributed to risk factors such as hyperten- authorities and the government to see the optimal trajectory of the
sion, diabetes and other cardiovascular diseases [11]. In Fig. 1a and implementation of non-pharmaceutical interventions.
1b we show the daily reported cases per day as of 18th June 2020. The rest of this paper is partition as follows: Section 2 con-
Since the outbreak of COVID-19, many mathematical models have tains the model formulation, definition of model parameters, in-
provided some insightful ideas to health authorities on how best variant region, boundedness, equilibrium points and stability anal-
to control the disease spread. For example, see the works of [13– ysis. Section 3 contains the model fit, parameter estimation, and
21]. Asamoah et al. [22] studied an (SEAQIs HRRADSp ) model to de- sensitivity analysis. Section 4 contains optimal control problem and
scribe COVID-19 human-human transmission dynamics for Ghana cost-effectiveness analysis. In Section 5, we give concluding re-
and Egypt; and obtained various epidemiological parameter values. marks and recommendations.
They suggested that: a continuous increase in the rate of diag-
noses, the rate of quarantine through doubling enhanced contact 2. Model formulation and analysis
tracing, and stringent safety measures in hospitals (and/or isola-
tion centres); with a constant supply of effective personal protec- This section discusses the transmission dynamics of COVID-19,
tive equipment’s (PPEs) will help reduce the control reproduction model analysis, properties and the equilibrium points, computa-
number Rc , to less than unity as the lockdown measures are been tion for the basic reproductive number and the stability analysis.
lifted. Adding natural recovery to their model, they showed that an In dealing with the dynamics of COVID-19 so as to implement
increase in natural recovery from the asymptomatic stage reduces the necessary control measures, we consider the compartments;
the control reproduction number. They also noticed that the choice Susceptible (S), Exposed/pre asymptomatic (E), Asymptomatic (A),
of a force of infection influences the control reproduction number. Symptomatic (I), Recovery (R) and the Virus in the Environment,
Although, human to human mathematical analysis has been estab- thus, on surfaces (V). As a deterministic model, we suppose that,
lished, not much mathematical analysis is done on the dynamics of the compartments depend on time. The total population of indi-
the virus in the environment and the role human play by releas- viduals, N(t) at time t is given as; N (t ) = S(t ) + E (t ) + A(t ) + I (t ) +
ing the viruses into the environment. The mathematical model that R(t ). We also consider that all parameters are positive and it is
studied the role of the environment focused on the seafood market only the I compartment which experiences disease induced death
in Wuhan and used Atangana-Bleanu-Caputo operators for their in- at a rate, α . The overall force of infection is given as λ = S[β (ηA +
vestigation without the use of optimal control model [23] (see also I ) + β1V ]. The model further assumes that, no exposed individual
[24]). The work by Zhang et al. [25] discussed the airborne trans- transmits the disease. The proportion of those in E class into both
mission as the main route for the spread of COVID-19. Ferretti et al. A and I classes is given as k2 (1 − γ )E and k1 γ E respectively. Sim-
[26], also quantified SARS-CoV-2 transmission dynamics and no- ilarly, individuals recovering from the A class is v2 (1 − φ )A and
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 3

Table 1
Parameters with their interpretation for model (1).

Parameter Description

 Recruitment rate
λ Force of infection
ω Natural death rate
β Transmission rate
β1 Transmission rate from the environment
η Relative transmissibility of asymptomatic individuals
γ Proportion of individuals who are timely diagnosis
k1 Progression rate from exposed to the symptomatic (severely infected) class
k2 Progression rate from exposed to the asymptomatic class
α Disease induced death rate
φ Proportion of asymptomatic patients who later move to the symptomatic (severely infected) class
v1 Progression from asymptomatic to the symptomatic (severely infected) class
v2 Progression from asymptomatic to the recovered class
The rate at which symptomatic (severely infected) patients recovers
ρ The rate at which the recovered individuals join the susceptible class
τ1 Natural decay rate of virus from the environment (Surfaces)
m1 The rate of viral release into the environment by asymptomatic patients
m2 The rate at which symptomatic patients release virus into the environment

Fig. 2. A model diagram of COVID-19 Dynamics.

those joining the I class from the A compartment is v1 φ A. Though, given that S ≥ 0, E ≥ 0, A ≥ 0, I ≥ 0, R ≥ 0 and V ≥ 0.
not much had been said with regards to the possibility of the
recovered individuals joining the susceptible population. All the
2.1. Analysis on the model
same we included it so to ascertain the impact of short and long-
term immunity on the dynamics of COVID-19, denoted as ρ , where
We simplify Eq. (1) to get the total differential equation as;
ρ ≥ 0. Table 1 further explains the individual parameters.
 dN
The differential equations of the COVID-19 model in Fig. 2 are
dt
=  − ωN − α I,
given below as; (2)
dV
dt
= m1 A + m2 I − τ1V,
dS
=  − ωS − β (IS + ηAS ) − β1V S + ρ R, where N = S + E + A + I + R.
dt
dE
= β (IS + ηAS ) + β1V S − k2 (1 − γ )E − k1 γ E − ωE, Theorem 1. The solution set {S(t), E(t), A(t), I(t), R(t), V(t)} is pos-
dt itive whenever the parameters are non-negative such that the initial
dA conditions are also given as
= k2 (1 − γ )E − ωA − v1 φ A − v2 (1 − φ )A, (1)
dt
dI {S(0 ) ≥ 0, E (0 ) ≥ 0, A(0 ) ≥ 0, I (0 ) ≥ 0, R(0 ) ≥ 0, V (0 ) ≥ 0},
= k1 γ E + v1 φ A − I − (ω + α )I,
dt The proof of Theorem 1 can be obtained using the procedures pre-
dR sented in [33], as shown below.
= v2 (1 − φ )A + I − ρ R − ωR,
dt
dV Proof. We let Y = (S, E, A, I, R, V )T and K0 = β I, K1 = β A, K2 = β1V
= m1 A + m2 I − τ1V, where T is transposition, then our differential Eq. (1) can be
dt
4 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

dY
rewritten in a matrix form as dt
= MY + M1 , where
⎛ ⎞
−(ω + K0 + ηK1 + K2 ) 0 0 0 ρ 0
⎜ (K0 + ηK1 + K2 ) −(K2 (1 − γ ) + k1 γ + ω ) 0 0 0 0 ⎟
⎜ 0 k2 ( 1 − γ ) −(ω + v1 φ + v2 (1 − γ )) 0 0 0 ⎟
M=⎜
⎜
⎟,
0 k1 γ v1 φ −( + ω + α ) 0 0 ⎟
⎝ ⎠
0 0 v2 ( 1 − φ ) − (ρ + ω ) 0
0 0 m1 m2 0 −τ1
⎛ ⎞

⎜0⎟
⎜0⎟
M1 = ⎜ ⎟
⎜ 0 ⎟.
⎝ ⎠
0
0

Now, using the third equation in model (1), thus

Which when solved gives
dA
= k2 (1 − γ )E − ωA − v1 φ A − v2 (1 − φ )A,
dt  
M2 (t ) = 1 − e(−ωt ) + N (0 )e(−ωt ) .
and deploying the method of integration factor and change of vari- ω
ables [33], yields Hence, by the comparison principle (see for instant Theorem 5 of
−(ω+v1 φ +v2 (1−φ ))t −(ω+v1 φ +v2 (1−φ ))t [38]), it accompanies that
A(t ) = A(0 )e +e k2 ( 1 − γ )
t
−(ω+v1 φ +v2 (1−φ ))s  
× E ( s )e ds, N (t ) ≤ 1 − e(−ωt ) + N (0 )e(−ωt ) . (5)
0 ω
t
= e−(ω+v1 φ +v2 (1−φ ))t A(0 ) + k2 (1 − γ ) E (s )e−(ω+v1 φ +v2 (1−φ ))s ds . Also from the second equation of (2), we let m = (m1 + m2 ) with
0 the assumption that 0 < A + I ≤ 
ω . Then,
Next, we consider the fourth, fifth, sixth equation of model (1) and dV m
using the same above process, we have ≤ − τ1V. (6)
dt ω
t t
I (t ) = e−( +ω+α )t I (0 ) + v1 φ A(s )e−( +ω+α )s ds + k1 γ E (s )e−( +ω+α )s ds . Now, let M3 to be a solution which is unique to the initial value
0 0
problem
t t d
R(t ) = e−(ρ +ω )t R(0 ) + v2 (1 − φ ) A(s )e−(ρ +ω )s ds + I (s )e−(ρ +ω )s ds . M3 (t ) = mω − τ1V for t > 0,
dt
0 0 (7)
M3 ( 0 ) = V ( 0 ).
t t
V (t ) = e−(τ1 )t V (0 ) + m1 A(s )e−(τ1 )s ds + m2 I (s )e−(τ1 )s ds . Which when solved gives
0 0
m  
Hence, from [34] and [35], we see that dA
| ≥ 0, dI
| ≥ 0, M3 (t ) = 1 − e(−τ1 t ) + V (0 )e(−τ t ) ,
dt t=t0 dt t=t0 ω
dR
|
dt t=t0
≥ 0, dV |
dt t=t0
≥ 0, and dE |
dt t=t0
≥ 0, provided E (t0 ) = 0, with
and, by the comparison principle (see for instant Theorem 5 of
same ideas being valid for S(t), this guarantees that the state vari-
[38]), it accompanies that
able remain positive during the entire scope of the study [33].
Grounded on this, we can say that matrix M has all its off diagonal m  
V (t ) ≤ 1 − e ( −τ t ) + V ( 0 ) e ( −τ t ) . (8)
entries to be non-negative and that matrix M1 ≥ 0, which proves ω
the properties of Metzler matrix [33,36]. Therefore it implies that
From (5) and (8) it connotes that, all the possible solution sets of
model (1) is positively invariant in R6+ [33]. For alternative proof to
the state variables S, E, A, I, R, V is bounded and, by the blow-up
the above assertion see Atangana et al. [37]. 
phenomena, hence the solutions exist and is defined for all t ≥ 0
[38]. Furthermore, for t → +∞, we get
2.2. Boundedness

0 ≤ N (t ) ≤ ,
Theorem 2. The model of the system in Eq. (1) has solutions bounded ω
∈ R6 written as m
within invariant region, 0 ≤ V (t ) ≤ .
ωτ1
 m
= (S, E, A, I, R, V ) ∈ R6+ : N (t ) ≤ , V (t ) ≤ .
ω ωτ1 This guarantees that, all the possible solution sets of the state vari-
ables S, E, A, I, R, V are contained in making the model both
where m = (m1 + m2 ).
mathematically and epidemiologically well posed (thus, model
Proof. From the first equation of (2) we have (1) is positively invariant). 
dN
≤ −ωN + . (3) 2.3. Disease free equilibrium point, E
dt
Now, take M2 to be a solution which is unique to the initial value
Here, we first focus on equilibrium points when there is no dis-
problem
d ease in the system. Considering Eq. (1), we put E = A = I = R =
dt
M2 (t ) =  − ωM2 (t ) for t > 0, V = 0. This indicates that, there is no disease in the system at
(4)
M2 ( 0 ) = N ( 0 ). this stage. Therefore, solving for the stationary points, we have
E = (S∗ , 0, 0, 0, 0, 0 ) where S∗ = 
ω.
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 5

2.4. Basic reproductive number, R0 which can be written as

R0 = RA + RI + RAe + RIe , (10)

For the R0 , we use the concept of the next generation approach.
Here, we seek to find the average number of new infections given where RA is the secondary infections generated by asymptomatic
that an infected individual is introduced into the population under persons through direct contact; RI is the secondary infections gen-
study [39]. We let G be the next generation matrix which consists erated by symptomatic persons through direct contact; RAe is the
of fi (x), v+
i
(x ) and v−
i
(x ), i = 1, 2, . . . , n ∈ N; where fi (x) is the rate secondary infection seeded by asymptomatic persons through the
at which a new infection occurs in compartment i. Also, v+ i
and environment; and RIe is the secondary infection seeded by symp-
v−i are the rate of immigration into compartment i and the rate tomatic persons through the environment. We can also express R0
at which new individuals are transferred from compartment i re- in terms of (T1 , C1 , Q1 , Y1 , X1 ) as;
spectively. We note that, all the functions are continuously differ-
entiable at least twice [39]. Let    C1 k1 γ τ1 +X1Y1 ητ1 +X1 v1 ρτ1  
m1 Q1 +C1 k1 γ m2 +X1 m2 v1 φ

 ∂ f (x )   ∂ V (x )  R0 = β τ + β τ .
i 0 i 0
ωT1 C Y
1 1 1
1 C Y
1 1 1

F= and V = ,
∂ xi ∂ xi ( x0 ) (11)

where Vi = v−
i
( x ) − v+
i
( x ). where T1 = γ k1 + k2 (1 − γ ) + ω, C1 = ω + φv1 + v2 (1 − φ ), Q1 =
Now the next generation matrix is defined as; k2 φv1 (1 − γ ), Y1 = α + + ω, X1 = k2 (1 − γ ).
G = F V −1 .
2.5. Endemic equilibrium point, E ∗∗
Hence, the R0 is given as the maximum absolute eigenvalue of the
next generation matrix (G) given that, σ contains all the eigenval- We let, E ∗∗ , be the endemic equilibrium points. Consider-
ues of G. This eigenvalue is known as the spectral radius (ρ ). This ing Eq. (1), we solve for the endemic equilibrium points E ∗∗ =
is represented as {S∗∗ , E ∗∗ , A∗∗ , I∗∗ , R∗∗ , V ∗∗ } with an assumption that; at the endemic
ρ (G ) = sup{|λ| : λ ∈ σ (G )}. state, E = 0, A = 0, I = 0, R = 0, V = 0. Hence, equating the deriva-
tives of Eq. (1) to zero, yields
Among the infected classes (E, A, I, V), we have fi as
⎡ ⎤  − ωS − β (IS + ηAS ) − β1V S + ρ R = 0,
β (AS + IS ) + β1V S
⎢ 0 ⎥ β (IS + ηAS ) + β1V S − k2 (1 − γ )E − k1 γ E − ωE = 0,
fi = ⎣ ⎦.
0 k2 ( 1 − γ )E − ω A − v1 φ A − v2 ( 1 − φ )A = 0, (12)
0 k1 γ E + v1 φ A − I − ( ω + α )I = 0,
Finding the Jacobian of the matrix fi gives v2 ( 1 − φ )A + I − ρ R − ω R = 0,
⎡ ⎤ ⎡ ⎤
0 ηβ S βS β1 S 0 ηβ 
ω β
ω β1 
ω
m 1 A + m 2 I − τ1 V = 0 .
⎢0 0 0 0 ⎥ ⎢0 0 0 0 ⎥
F =⎣
0 ⎦ ⎣0 ⎦.
= Now, expressing the other state variables in terms of E, it implies
0 0 0 0 0 0
that
0 0 0 0 0 0 0 0
X1 E
Considering the same compartments (E, A, I, V ), we get the matrix A= ,
C1
V as
⎡ ⎤ k1 γ E v1 Q1 E
k2 ( 1 − γ )E + k1 γ E + ω E I= + ,
Y1 Y1C1
⎢−k2 (1 − γ )E + ωA + v1 φ A + v2 (1 − φ )A⎥
Vi = ⎣
−k1 γ E − v1 φ A + I + (ω + α )I ⎦. v2 (1 − φ )X1 E k1 γ E Q1 E
R= + + ,
−m1 A − m2 I + τ1V (ρ + ω )C1 (ρ + ω )Y1 Y1C1
m1 X1 E  
Finding the Jacobian matrix of Vi gives V = + m2 kY1 γτ E + Y QC1 Eτ .
⎡ ⎤ τ1C1 1 1 1 1 1

γ k1 + k2 ( 1 − γ ) + ω 0 0 0
⎢ −k2 (1 − γ ) ω + φv1 + v2 (1 − φ ) 0 0⎥ We now substitute A, I and V into second Eq. (12) and factorizing
V =⎣
0⎦
.
−γ k 1 −φv1 α+ +ω E out gives
0 −m1 −m2 τ1   k1 γ     
βS Y1
+ Q1
Y1 C1
+ ηCX11 + β1 S mτ11CX11 + m2 Yk11τγ1 + Q1
Y1 C1 τ1
− T1 E = 0.
After computing for the eigenvalues of the matrix G, we have that
the maximum absolute eigenvalue, R0 is given as; (13)
βηκ2 (1 − γ ) From the initial hypothesis, E = 0 and this implies that,
R0 =
ω (κ2 (1 − γ ) + κ1 γ + ω )(ω + ν1 φ + ν2 (1 − φ ))  k1 γ    
  
secondary infection seeded by A state through direct contact
β S∗ Y1
+ Q1
Y1 C1
+ ηCX11 + β1 S∗ mτ11CX11 + m2 Yk11τγ1 + Q1
Y1 C1 τ1
− T1 = 0.
β [(ω + ν1 φ + ν2 (1 − φ ))κ1 γ + κ2 ν1 (1 − γ )φ ] (14)
+
ω (κ2 (1 − γ ) + κ1 γ + ω )(ω + ν1 φ + ν2 (1 − φ ))(ε + ω + α )
   Making S∗ in Eq. (14) the subject gives
secondary infection seeded by I state through direct contact
T1
β1 m1 κ2 (1 − γ ) S∗ =  k1 γ     . (15)
+
ω (κ2 (1 − γ ) + κ1 γ + ω )(ω + ν1 φ + ν2 (1 − φ ))τ1 β Y1
+ Q1
Y1 C1
+ ηCX11 + β1 mτ11CX11 + m2 Yk11τγ1 + Q1
Y1 C1 τ1
  
secondary infection seeded by A state through the environment Now, adding first equation and second equation of (12), we substi-
β1 m2 [(ω + ν1 φ + ν2 (1 − φ ))κ1 γ + κ2 ν1 (1 − γ )φ ] tute S∗ from Eq. (15) and simplify E∗ we get
+ ,
ω (κ2 (1 − γ ) + κ1 γ + ω )(ω + ν1 φ + ν2 (1 − φ ))(ε + ω + α )τ1 T1 
   E∗ = − , (16)
secondary infection seeded by I state through the environment HL L
(9)
6 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

where From the set above, the equilibrium y∗ is globally stable if ζ con-
   m1 X1  k1 γ  stitutes only the equilibrium y∗ [41](a fractional prove of stability
k1 γ Q1 ηX1
H=β + + + β1 τ1C1 + m2 Y1 τ1
+ Q1
Y1 C1 τ1
, can be extracted from [42]).
Y1 Y1C1 C1
ρv2 (1 − φ )X1 ρ k1 γ ρ Q1 Theorem 4. The disease-free equilibrium point of the COVID-19
L= + + − T1 . model is globally stable if R0 < 1.
(ρ + ω )C1 (ρ + ω )Y1 (ρ + ω )Y1C1
But, we know that the basic reproduction number is expressed Proof. We employ the approach in [43] to analyze both the stabil-
as ity at disease free and endemic equilibrium. We define a Lyapunov,
⎡   ⎤ L for the disease-free equilibrium point as follows;
C1 k1 γ τ1 + X1Y1 ητ1 + X1 v1 ρτ1 !
 ⎢ β βη βv1 φ
β1 m2 v1 φ β1 m1
 τ1C1Y1 ⎥.
ωT1 ⎣+β m1 Q1 + C1 k1 γ m2 + X1 m2 v1 φ ⎦
R0 = L=E+ + + + A
C1 Y1C1 Y1 τ1C1 C1 τ1
1
τ1C1Y1 !
β β1 m2 β1
Therefore, we express the endemic equilibrium points, (S∗ , E∗ , A∗ , + + I+ V.
Y1 Y1 τ1 τ1
I∗ , R∗ , V∗ ) in terms of R0 as;
Differentiating L with respect to t gives;
 ! !
S∗∗ = , A∗∗ = g1 E ∗∗ , I∗∗ = g2 E ∗∗ ,
R0 ω βη βv1 φ β1 m2 v1 φ β1 m1 ˙ β1 m2 ˙ β1 ˙
!! L˙ = E˙ +
C1
+
Y1C1
+
Y1 τ1C1
+
C1 τ1
A+
Y1 τ1
I+
τ1
V.
∗∗ ∗∗ ∗∗ m1 X1 k1 γ Q1
R = g3 E , V = + m2 + E ∗∗ ,
τ1C1 Y1 τ1 Y1C1 τ1 We substitute E˙ , A˙ , I˙, V˙ from Eq. (1) into L˙ gives;
  1

E ∗∗ = 1− . L˙ = β (IS + ηAS ) + β1V S − k2 (1 − γ )E − k1 γ E − ωE
ω (g1 + (1 − α )g2 + g3 + 1 ) R0 !
   βη βv1 φ β1 m2 v1 φ β1 m1
k1 γ v2 (1−φ )X1 k1 γ + + + +
where g1 =
X1
C1 , g2 = Y1 +
v1 Q1
Y1 C1 , g3 = (ρ +ω )C1 + (ρ +ω )Y1 +
C1 Y1C1 Y1 τ1C1 C1 τ1
  
Q1
Y1 C1 . Therefore, the endemic equilibrium for the COVID-19 model × k2 ( 1 − γ )E − ω A − v1 φ A − v2 ( 1 − φ )A
remains positive if and only if R0 > 1, since at the endemic !
equilibrium, E > 0, R > 0, I > 0, A > 0. Fulfilling, that we have a β β1 m2  
+ + k1 γ E + v1 φ A − I − ( ω + α )I
unique endemic equilibrium point E ∗∗ , when R0 > 1. (see [40] for Y1 Y1 τ1
an alternative prove on unique endemic equilibrium point. β1  
+ m 1 A + m 2 I − τ1 V .
τ1
2.6. Global stability
After further simplification we have
We focus on the application of Lyapunov’s function in this area.       
Lyapunov function is a scalar function which may be employed to L˙ = T1 ωT1 β ηCX11 + γYk11 + Q1
Y1 C1
+ β1 m1 X1
τ1 C1 + mτ21γY1k1 + τm1 Y21QC11 −1
ascertain global stability [41]. = T1 (R0 − 1 )E
Definition 1. Let the equilibrium of y = f (y ) be such that f : y∗ Therefore, L˙ ≤ 0 whenever R0 < 1. Also, L˙ = 0 if and only if E = A =
Rn → Rn . We define L to be a continuous scalar function, implying; I = V = 0. We then define the invariant set as
L : Rn → R. If the conditions
ζ = {(E, A, I, V ) ∈ Rn : L˙ (y ) = 0}.
L ( y∗ ) = 0, L (y ) > 0 ∀y = y∗ ,
Hence, by the Krasovkii-LaSalle theorem, it follows that since ζ
are satisfied, then the function L is said to be positive definite. houses only the equilibrium E ∗ , the E ∗ is said to be globally stable
Also, a scalar function L(y) such that L : Rn → R is called radially whenever R0 < 1 [41]. 
unbounded if L(y)→∞ as ||y||→∞ [41].
Theorem 5. If R0 > 1, the endemic equilibrium point of the COVID-
Based on the preliminary notes above, we now state the Lya- 19 model is globally asymptotically stable in [44].
punov stability theorem.
Proof. Similarly, to check for the global stability for the endemic
Theorem 3 (Lyapunov Stability Theorem). The equilibrium, y∗ is equilibrium point, we define a Lyapunov function, L1 ;
globally stable if the function, L(y) is radially unbounded and positive  S
  E

definite globally such that it has globally negative time derivative, L1 = S − S∗∗ − S∗∗ ln + h1 E − E ∗ − E ∗∗ ln
S∗∗ E ∗∗
L˙ (y ) < 0 ∀y = y∗ .    
A I
+ h2 A − A∗∗ − A∗∗ ln + h3 I − I∗∗ − I∗ ln ∗
We say that; the function L(y) is a Lyapunov function if it satisfies the A∗∗ I
above theorem, the proof can be found in [41].  R
  V

+ h4 R − R∗∗ − R∗∗ ln ∗∗ + h5 V − V ∗∗ − V ∗∗ ln ∗∗ .
Another important theorem which also plays a key role here is R V
the Kransovkii-LaSalle theorem. This is an extension of Lyapunov Differentiating the function above gives;
function. In summary, this theorem puts forward that; consider-  S∗∗
 
E ∗∗

A∗∗ ˙
I∗∗
  
ing an autonomous system, y = f (y ) which has equilibrium, y∗ L˙ 1 = 1 − S˙ + h1 1 − E˙ + h2 1 − A + h3 1 − I˙
and that f (y∗ ) = 0, we assume there is a continuously differen- S E A I
 R∗∗
  V ∗∗

tiable positive definite and radially unbounded function L : Rn → R
+ h4 1 − R˙ + h5 1 − V˙ . (18)
which meets the condition L˙ (y ) ≤ 0 ∀ t, y ∈ Rn . We then define the R V
invariant set as
Substituting Eq. (1) into Eq. (18) with further simplification
ζ = {y ∈ Rn |L˙ (y ) = 0}. gives;
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 7

 1
  1
 x2 x3 x3
L˙1 = −ω 1 − 2
S+ 1− + x4 − x4 x1 β I∗∗ S∗∗ −h3 k1 γ E ∗∗− h3 v1 φ A∗∗ − h4 v2 (1 − φ )A∗∗
x1 x1 x4 x4 x5
 1
 ∗∗ x4 ∗∗ x3 ∗∗ x4
+ 1− + x3 − x3 x1 βηA∗∗ S∗∗ − I h4 − h5 m1 A − h5 m2 I .
x1 x5 x6 x6
 1
  1 x5
 Considering the expression
+ 1− + x6 − x6 x1 β1V ∗∗ S∗∗ + 1 − + x5 − ρ R∗∗

x1

x1 x1 β I∗∗ S∗∗ h1 = β I∗∗ S∗∗ ,
x1 x4
+ x1 x4 − x2 − + 1 β I∗∗ S∗∗ h1 we have that, h1 = 1. This implies that the coefficients of x1 x4 , x3 x1
x2
 x1 x3
 and x6 x1 are all 0. Equating the coefficients of x2 , x3 , x4 , x5 and x6
+ x1 x3 − x2 − + 1 βηA∗∗ S∗∗ h1 to 0 and solving for h2 , h3 , h4 and h5 gives;
x2
  βηA∗∗ S∗∗ (β I∗∗ S∗∗ + βηA∗∗ S∗∗ + β1V ∗∗ S∗∗ )v1 φ A∗∗
x1 x6 h2 = +
+ x1 x6 − x2 − + 1 β1V ∗∗ S∗∗ h1 k2 (1 − γ )E ∗∗ k1 γ E ∗∗ k2 (1 − γ )E ∗∗
x2 v2 (1 − φ )A∗ ρ R∗∗ m1 A∗∗ β1V ∗∗ S∗∗
 x2
 + + ,
+ 1 + x2 − x3 − h2 k2 (1 − γ )E ∗∗ (v2 (1 − φ )A∗∗ + I∗∗ )k2 (1 − γ )E ∗∗ (m1 A∗∗ + m2 I∗∗ )k2 (1 − γ )E ∗∗
x3 β I S + βηA S + β1V S
∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗
    h3 = ,
x2 x3 k1 γ E ∗∗
+ 1 + x2 − x4 − h3 k1 γ E ∗∗ + 1 + x3 − x4 − h3 v1 φ A∗∗
x4 x4 ρ R∗∗
 x3
 h4 =
v2 (1 − φ )A∗∗ + I∗∗
,
+ 1 + x3 − x5 − h4 v2 (1 − φ )A∗∗ β1 V S
∗∗ ∗∗
x5 h5 = .
 x4
 m1 A∗∗ + m2 I∗∗
+ 1 + x4 − x5 − I∗∗ h4
x5 Therefore, T can be rewritten as
 x3
  x4
  
1 x1 x4
+ 1 + x3 − x6 − h5 m1 A∗∗ + 1 + x4 − x6 − h5 m2 I∗∗ T = −β I∗∗ S∗∗ x2 − x4 + + −2
x6 x6 x1 x2
 1 x3 x1

where x1 = S E A I R V −βηA∗∗ S∗∗ x2 − x3 + + −2
S∗∗ , x2 = E ∗∗ , x3 = A∗∗
, x4 = I∗∗ , x5 = R∗∗ and x6 = V ∗∗ . x1 x2
We let  1 x1 x6
 1
 x5

 1
 −β1V ∗∗ S∗∗ x2 − x6 + + − 2 − ρ R∗∗ − 1 + − x5 + −2
x1 x2 x1 x1
L˙ 1 = −ω 1− 2
S + T ( x1 , x2 , x3 , x4 , x5 , x6 ),    
x1 x2 x2
−h2 k2 (1 − γ )E ∗∗ x3 − 1 − x2 + − h3 k1 γ E ∗∗ x4 − 1 − x2 +
x3 x4
where    
x3 x3
−h3 v1 φ A∗∗ x4 − 1 − x3 + − h4 v2 (1 − φ )A∗∗ x5 − 1 − x3 +
T = β I S + βηA S + β1V S + ρ R + β I S h1
∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗
x4 x5
   
+βηA∗∗ S∗∗ h1 + h2 k2 (1 − γ )E ∗∗ + h3 k1 γ E ∗∗ − I∗∗ h4 x5 − 1 − x4 +
x4
− h5 m1 A∗∗ x6 − 1 − x3 +
x3
x5 x6
+h3 v1 φ A∗∗ + β1V ∗∗ S∗∗ h1 + h4 v2 (1 − φ )A∗∗ + I∗ h4  
x4
+h5 m1 A∗∗ + h5 m2 I∗∗ −h5 m2 I∗∗ x6 − 1 − x4 + .
x6
 1
+ − β I∗∗ S∗∗ − βηA∗∗ S∗∗ − β1V ∗∗ S∗∗ − ρ R∗∗ It then follows that, T ≤ 0 if x1 = 1, x2 = 1, x3 = 1, x4 = 1, x5 = 1
x1
  and x6 = 1. Hence we may conclude that;
+ − β I∗∗ S∗∗ h1 − βηA∗∗ S∗∗ h1 − β1V ∗∗ S∗∗ h1 + h3 k1 γ E ∗ x2  1

L˙ 1 = −ω 1 − 2
S + T ( x1 , x2 , x3 , x4 , x5 , x6 ) ≤ 0.
 x1
+ βηA S − h2 k2 (1 − γ )E + h3 v1 φ A + h4 v2 (1 − φ )A
∗∗ ∗∗ ∗∗ ∗ ∗∗
By LaSalle theorem, the invariant set is defined as

+h5 m1 A∗∗ x3 ζ1 = {(S, E, A, I, R, V ) ∈ D : L1 = 0}.
  Since the invariant set, ζ 1 only contains the endemic equilibrium
+ β I∗∗ S∗∗ − h3 k1 γ E ∗∗ − h3 v1 φ A∗∗ + I∗ h4 + h5 m2 I∗ x4 (S∗∗ , E∗∗ , A∗∗ , I∗∗ , R∗∗ , V∗∗ ), then the endemic equilibrium is said to
  be globally asymptotically stable under the given region . 
+ ρ R∗∗ − h4 v2 (1 − φ )A∗∗ − I∗∗ h4 x5
3. Parameter estimation and numerical simulations
 
+ − h5 m1 A∗∗ − h5 m2 I∗∗ + β1V ∗∗ S∗∗ x6 In this section, our focus is to verify the validity of the model.
  This is achieved by fitting and comparing the proposed model with
+ − β I∗∗ S∗∗ + β I∗∗ S∗∗ h1 x4 x1 a real data to know its degree of accuracy. This will justify the
  model’s competency on predicting for a realistic outcome. Consid-
+ − βηA∗∗ S∗∗ + βηA∗∗ S∗∗ h1 x3 x1 ering Ghana as a case study, the cumulative daily reported cases
of COVID-19 were extracted from the situation reports of the WHO
  and the Ministry of Health, Ghana [11]. The collected data ranges
+ − β1V ∗∗ S∗∗ + β1V ∗∗ S∗∗ h1 x6 x1
from the onset of the pandemic in Ghana; that is from 12th March
x5 x1 x4 x3 x1 to 7th May, 2020. As of 7th May, 2020, 4012 cases were confirmed
−ρ R∗∗ − β I∗∗ S∗∗ h1 − βηA∗∗ S∗∗ h1 with 18 deaths and 323 recovered individuals. No clear detail was
x1 x2 x2
x6 x1 given with respect to the relapse of the recovered individuals into
∗ x2
−β1V S h1
∗∗ ∗∗
− h2 k2 ( 1 − γ ) E the susceptible class though this study factored. Estimating param-
x2 x3
eter values is very key as far as epidemiological study is concerned.
8 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Table 2 3.1. Basic reproduction number, R0 with no control

Estimated parameter values for the COVID-19 model (1).

Parameter Range Baseline value Reference Our main objective here is to compute the basic repro-
η 0.62811041 − 0.6366 day
−1
0.6323 Estimated
duction number of the model (1). Using the values η =
ω 0.0 0 0 042578 day
−1
0.000042578 [11] 0.62811041, ω = 0.0 0 0 042578, ρ = 0.41138431, β = 6.038 × 10−8 ,
ρ 0.41138431 day
−1
0.41138431 Estimated β1 = 4.00199 × 10−8 , γ = 0.010 0 0 0 01,  = 1319.294, k1 = 0.07142
β 6.038 × 10−8 − 0.8038 6.038 × 10−8 Estimated 238, k2 = 0.14285824, α = 0.0 0440 0 0 0, φ = 0.0 050 0 0 05, v1 = 0.20
γ 0.010 0 0 − 0.0648 0.0374 Estimated 005051, v2 = 0.79999398, = 0.0805840, τ1 = 0.290 0 0 0 0, m1 =
β1 4.00199 × 10−8 4.00199 × 10−8 Estimated
−1 0.01780400, m2 = 0.92152716, and substituting it into the R0
 1319.294 day 1319.294 [45,46]
k1 0.19230 0 0 − 0.07142238 day
−1
0.1318 Estimated
expression result in;
k2 0.176300 − 0.14285824 day
−1
0.1596 Estimated    C1 k1 γ τ1 +X1Y1 ητ1 +X1 v1 ρτ1   
α −1 R0 = β τ + β1 m1 Q1 +C1 kτ11γCm 2 +X1 m2 v1 φ
0.0044 − 0.0099 day 0.0072 Estimated ωT1 1 C1 Y1 1 Y1
φ 0.00500005 0.0050 Estimated
1319.294
 
v1 0.1929 − 0.20 0 0 day
−1
0.1965 Estimated = 7.3634285 × 10−9 + 1.7708567 × 10−9
−1 4.2578 × 10−5 × (0.1421865 )
v2 0.79999398 day 0.8000 Estimated
0.0805084 − 0.098026529 day
−1
0.0893 [11] = 1.99,
−1
τ1 0.290 0 0 − 0.3333 day 0.3117 [47]
−1 where T1 = 0.14218646, C1 = 0.7970368, Q1 = 0.0 0 01415, Y1 =
m1 0.01780400 day 0.0178 Estimated
m2 0.92152716 day
−1
0.9215 Estimated 0.0849510, X1 = 0.1414297. However, using the baseline values in
Table 2 and substituting it into the R0 expression result in;
1319.294
 
R0 = 1.05604 × 10−8 + 6.71352 × 10−9
4.2578 × 10−5 × (0.15860 )
= 3.37,

where T1 = 0.15860, C1 = 0.79703, Q1 = 0.0 0 015, Y1 = 0.09654,

X1 = 0.15363. Therefore, the average basic reproduction number,
Ra0 , without control strategies in Ghana is approximately 2.68. This
indicates that on average a single infected individual can spread
the disease to 2 to 3 susceptible individuals. Hence, strengthen-
ing prophylaxis such as social distancing, wearing of nose masks,
regular hand washing with soap and contact tracing, will highly
perturb the key parameters aiding the spread of the disease. Sup-
pose that, the implementation of these measures decreases the es-
timated values of η, β and β 1 to 0.32811041, 4.038 × 10−8 and
2.00199 × 10−8 respectively, then the basic reproduction number
with these control outcomes will be given as 0.79774. This shows
how effective these control measures are in COVID-19 battle field.

Fig. 3. Fitted diagram of confirmed cases in Ghana and the proposed model. 3.2. Sensitivity analysis

Varying estimated parameter values might give different output

We utilize the least squares method in this context since it is very as seen in 3.1 and it is on this note that, we want to explore pa-
efficient and reliable [41]. This method seeks to fit the observed rameters which significantly influence the end results of the model
data sets, Yi , with the estimated values, Xi , such that; the sum of under study. This concept is very important since it discusses and
squares of errors between the observed and fitted curve is minimal informs us about the most relevant factors to deal with so as to
[41]. The sum of squares error, SSE, is illustrated mathematically minimize the spread of COVID-19.
as;
Definition 2. The parameter, p is said to be sensitive if any small
"
k alteration on p causes a significant change in the solution. It is
SSE = (Yi − Xi )2 . worthy to note that, the parameter p is termed to be locally sensi-
i=1 tive if the change in the value of parameter p influences the output
of the model. In the same way, global sensitivity takes into account
Table Appendix A account for the daily recorded cases and daily
the overall change in the model output as a result of the change in
percentage change in the number of recorded cases.
all parameter values within their respective range [41].
The estimated parameter values and the fitted model using
p
Python programming language are shown in Table 2 and Fig. 3 re- In computing for the normalized sensitivity index ( R ) on the
0
spectively. The blue points in Fig. 3 represent the cumulative num- R0 for each of the parameters p, we use the formulae below [41];
ber of daily confirmed cases while the red line designates the fit-
p ∂ R0 p
ted model. From the estimated parameter values in Table 2, it is = .
clear that; even if all the infected individuals are treated, the sus-
R0
∂ p R0
ceptible class is still vulnerable to be infected since the rate at Applying the formula above gives the parameters with their sensi-
which the environment contribute to infection, β 1 , is significant, tivity index in the Table 3.
thus 4.00199×10−8 . Based on the estimated value for γ , only 1% From Table 3, all parameters with positive values, η, β , β 1 , γ ,
of the exposed individuals join the infected compartment without , k1 , φ , v1 , m1 and m2 indicate the increase in transmission of
passing through the asymptomatic state. Again, only 0.5% of the in- COVID-19 in the population in accordance to an increase in these
dividuals in the asymptomatic class join the symptomatic class at parameter values. Prominent among these are , β and η. On the
the rate of 0.2 per day while the rest of the asymptomatic popu- other hand, parameters with negative sensitivity index, ω, k2 , α , v2 ,
lace recover at the rate of 0.799994 per day. and τ 1 indicate the decrease in the spread of COVID-19 for each
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 9

Table 3
Normalized sensitivity index for each parameter for the COVID-19 model (1).

Parameter Description Sensitivity Index

η Relative transmissibility by asymptomatic individuals +0.7367

ω Natural death rate -1.0005
β Transmission rate by asymptomatic and symptomatic individuals +0.8060
β1 Environmental transmission +0.1939
γ Proportion of individuals who are timely diagnosis +0.1693
 Recruitment rate +1.0 0 0 0
k1 Progression rate from exposed to the symptomatic class +0.1676
k2 Progression rate from exposed to the asymptomatic class −0.1673
α Disease induced death rate −0.0112
φ Proportion of A patients who later move to the I class +0.0460
v1 Progression from asymptomatic to the symptomatic class +0.0419
v2 Progression from asymptomatic to the recovered class −0.8263
The rate at which symptomatic patients are recovered −0.2043
τ1 Natural decay rate of virus from the environment (Surfaces) −0.1939
m1 The rate of viral release into the environment by asymptomatic patients +0.0477
m2 The rate at which I patients release virus into the environment +0.1461

So, does the recruitment rate, . Therefore, the government’s de-

cision on travel restriction has really been of help to the coun-
try against the spread of the disease since this contributed mas-
sively to reduce the secondary infections which might have been
contracted as a result of individuals from highly infected coun-
tries. Also, the analysis figures out that, mass testing and proper
observance of social distancing with its complementary measures
would far help in this fight. This analysis also establishes the fact
that; adapting to healthy living/lifestyle which seek to boost the
immune system such as proper dietary habit and abstinence from
alcohol would significantly increase the number of individuals in
the recovery compartment. One key feature which needs much at-
tention is the rate at which viruses are released into the environ-
ment. This rate is very high among the asymptomatic patients, see
Fig. 4. This can be attributed to the fact that; since individuals in
this class are mostly unaware about their contraction of COVID-19,
strict measures against the virus’ propagation is not properly ob-
Fig. 4. Global sensitivity plot . served among individuals in this class contributing to this effect.
Fig. 5a, is the simulation of the situation as of 7th May, 2020. Here,
the basic reproduction number, R0 is 1.99. The graph confirms this
unit increase in these parameters. Exploiting sensitivity analysis on figure that, if no effective control measures are employed, then
this study enlightens policy makers on where to put more efforts this disease would continue to be experienced in the population
in the fight against COVID-19. For example, the sensitivity index of for all time until the population gets diminished. Fig. 5b describes
 on R0 is 1. This implies that, a 1% change in the value of  will the projections of the infected and asymptomatic individuals with
cause a corresponding increase in R0 by 1%. In the same manner, no control measure for about two years period. It clearly depicts
a 1% change in β increases the number of secondary infections by that, the number of individuals in the asymptomatic compartment
0.8%. From the global sensitivity analysis as shown in Fig. 4, it is far exceeds that of the infected class even as time transcends. This
clear that, fortifying measures which will aid in accelerating , γ confirms the results of the parameter estimates that, only 1% of all
and τ 1 will reduce the R0 . These measures may include fumigat- the exposed individuals branch into the symptomatic class at the
ing the environment (surfaces) especially in crowded areas such rate of 0.1676 while the rest move to the asymptomatic class. The
as markets and essential service providing institutions like banks, model shows that, asymptomatic individuals can join the severely
police stations and hospitals. Regular disinfecting of the transport infected class. However, the estimated parameter value shows that,
systems would also increase the virus diminishing rate from the only few people experience such situation, that is, 0.5%.
system, enhancing the outcome of the control reproductive num- Though, not much had been said with regards to the possibil-
ber. Though, ω contributes significantly towards the reduction in ity of the recovered individuals joining the susceptible population.
the R0 , but it is not considered as good candidate to be used as Notwithstanding, after our model’s fitting, it was shown that, this
−1
a control variable. This is grounded on the fact that, ω, is the nat- situation might be possible at the rate of 0.41138 day as shown
ural death rate of the population of which we do not have much in Fig. 8b. This calls attention that, treatment must be geared to-
dominance. wards provision of long term or permanent immunity so as to
In the same setting as illustrated in Fig. 4, the parameters m2 , break the path between the recovered class and reinfection.
m1 , k2 , , β 1 , β , η prove to be more influential in increasing the
reproduction number after performing the global sensitivity analy-
sis. This is backed by the results in Figs. 5, 6, 7 and Fig. 8 where 4. Optimal control on the model
the effects of β , β 1 , m1 , m2 , η and ρ on the infected population
are explored. From the Fig. 5a, 5 b, 6 a, 6 b, 7 a, 7 b, 8 a, 8 b, Considering the system x (t ) = f (x(t )) with x(0 ) = x0 , such
we have a confirmation that, the number of the infected individ- that x0 ∈ Rn where f : Rn → Rn and x : [0, ∞ ) → Rn . We introduce
uals increase as the parameters, β , β 1 , m1 , m2 , η and ρ increase. variables responsible for the control ui , i = 1, 2, 3 . . . n ∈ N. We then
10 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Fig. 5. The dynamics of COVID-19 in Ghana considering the impact of the environment.

Fig. 6. The effects of varying β , β 1 on the infected population.

Fig. 7. The effects of varying m1 and m2 on the infected population.

 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 11

Fig. 8. The effects of η and ρ on the infected population.

have; subject to the constraints;

x (t ) = f (x(t ), u(t )), x(0 ) = x0 ,
dS
= (1 − u1 ) − ωS − (1 − u2 )β (IS + ηAS ) − (1 − u2 − u3 )β1V S,
with u : [0, ∞ ) → A ⊂ Rn . The admissible control set is given as; dt
U ∗ = {u(t ) ∈ L (t0 , t f )|u(t ) ∈ A}. The aim here is to target the best dE
control variables, ui , which can efficiently reduce the rate of sec- = (1 − u2 )β (IS + ηAS ) + (1 − u2 − u3 )β1V S − k1 γ E
dt
ondary transmission at a minimum cost of their implementation at −k2 (1 − γ )E − ωE,
any time t(0) ≤ t ≤ t(f) [41]. That is, we seek to achieve a reduction dA
in the number of individuals in the susceptible, exposed, asymp- = k2 (1 − γ )E − ωA − v1 φ A − v2 (1 − φ )A,
dt
tomatic, severely infected classes and also reduce the content of
dI
the virus in the system at a minimum cost simultaneously. To = k1 γ E + v1 φ A − I − (ω + α )I,
dt
achieve the above objective demands a lot of constructive consider-
dR
ations. For example, implementing total lock down for about two = v2 (1 − φ )A + I − ωR,
dt
months as a control strategy in this context might highly prove
dV
not to be feasible. The loop hole here is; is the country adequately = (1 − u4 )m1 A + (1 − u4 )m2 I − (u5 + τ1 )V. (20)
prepared both financially and technically to provide to the satis- dt
faction of its inhabitants the basic needs such as food, water and Where, S > 0, E ≥ 0, A ≥ 0, I ≥ 0, R ≥ 0, V ≥ 0. From Eq. (19),
others throughout the whole period assigned for this measure? It we assume that, the weight constant of the exposed, infected
is highly probable the answer might be a big no. It is an undeni- (A and I classes) and the virus in the system is 1. Also, to better ob-
able fact that, this measure might prove impractical in controlling serve and understand the influence of these control strategies on
the spread in this country. This is why there is a need to objec- the model, we assumed that; no recovered individual is vulnera-
tively sort for more dense restrictive measures with flexible and ble to be reinfected in Eq. (20). We accounted for the respective
feasible approaches to be employed in this setting so as to control affiliated costs, b1 u21 , b2 u22 , b3 u23 , b4 u24 , b5 u25 , which are possible to
the disease. We rely on the Pontryagin’s maximum principle as ap- be incurred during implementation where the square denotes their
plied in [48] for this analysis. We base on the premises above to severity.
set below likely control strategies: It is very necessary to ensure that, the proposed optimal solu-
1. u1 = The effective testing and quarantine when boarders are tion exists. For this reason, we employ Filippove-Cesari theorem as
opened. used in [37]. In this case, we show that, the existence of the opti-
2. u2 = Intensifying the usage of nose masks and face shields mal control solution is guaranteed if the following conditions are
through education. satisfied;
3. u3 = Cleaning of surfaces with home-based detergents.
1. The admissible control set is compact and bounded.
4. u4 = Safety measures adopted by the asymptomatic and symp-
2. The control sets together with the state variables is non-empty.
tomatic individuals such as; practising proper cough etiquette
3. A linear function in the state and control variables bound the
(maintaining a distance, cover coughs and sneezes with dis-
state systems of differential equations.
posable tissues or clothing and wash hands after cough or
4. The convexity of the integrand of cost functional with respect
sneezes).
to u on the set A [37].
5. u5 = Fumigating commercial areas such as markets.
The objective functional under discussion, Q, which is to be We now have the Hessian matrix of the given cost functional
minimized is given as; as;
tf ⎡b 0 0 0 0
⎤
Q (u1 , u2 , u3 , u4 , u5 ) = min 1
t0 ⎢0 b2 0 0 0⎥
  ⎢0 0 b3 0 0⎥
× E +A+I +V +
b1 2 b2 2 b3 2 b4 2 b5 2
u + u + u + u + u dt; ⎣ ⎦
2 1 2 2 2 3 2 4 2 5 0 0 0 b4 0
0 0 0 0 b5
(19)
12 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Since the computed Hessian matrix above is everywhere posi- The optimal control strategies with respect to the befitting varia-
tive definite, it follows that, the objective functional, Q(u1 , u2 , tion argument is given as;
u3 , u4 , u5 ) is strictly convex [37]. We also have that, ∃ b =  ! #
min{b1 , b2 , b3 , b4 , b5 } > 0 given that the integrand of the objective λ1 
u1 = min max 0, , u1max ;
functional, b1
 ! #
b1 2 b2 2 b3 2 b4 2 b5 2 (−λ1 + λ2 )[β (I S + nA S ) + β1V  S ]
Q = E +A+I +V + u + u + u + u + u , u2 = min max 0, , u2max ;
2 1 2 2 2 3 2 4 2 5 b2
b1 2 b2 2 b3 2 b4 2 b5 2  ! #
≥ u + u + u + u + u , (−λ1 + λ2 )β1V  S
2 1 2 2 2 3 2 4 2 5 u3 = min max 0, , u3max ;
b3
≥ b( u 1 + u 2 + u 3 + u 4 + u 5 ) ,
2 2 2 2 2
 ! #
λ6 [m1 A + m2 I ]
u4 = min max 0, , u4max ;
holds under the condition E + A + I + V ≥ 0. Applying Pontryagin’s b4

maximum principle where we have the state variables as S = S ,  ! #
    
E = E , A = A , I = I , R = R and V = V gives the Hamiltonian λ6V 
u5 = min max 0, , u5max .
function; b5

b1 2 b2 2 b3 2 b4 2 b5 2 We progress with the numerical simulations on the optimal control

H = E  + A + I  + V  + u + u + u + u + u by using the estimated parameters in Table 2.
2 1 2 2 2 3 2 4 2 5
+λ1 [(1 − u1 ) − ωS − (1 − u2 )β (I S + ηA S )
−(1 − u2 − u3 )β1V  S ] 4.1. Numerical results of the optimal control analysis

+λ2 [(1 − u2 )β (I S + ηA S ) + (1 − u2 − u3 )β1V S

     
Using the forward-backward sweep method as introduced by
−k1 γ E  − k2 (1 − λ )E  − ωE  ] Lenhart and Workman [49]; and extensively explored by [28,50–
+λ3 [k2 (1 − γ )E  − ωA − v1 φ A − v2 (1 − φ )A ] 52] and the references therein. We considered the following
equal weight factors, initial data for the state variables along
+λ4 [k1 γ E  + v1 φ A − I − (ω + α )I ]
with the parameter values, thus b1 = 10, b2 = 10, b3 = 10, b4 =
+λ5 [v2 (1 − φ )A + I − ωR ] + λ6 [(1 − u4 )m1 A 10, b5 = 10; S(0 ) = 304160 0 0, E (0 ) = 5, A(0 ) = 5, I (0 ) = 2, R(0 ) =
+(1 − u4 )m2 I − (u5 + τ1 )V  ]. 0, V ( 0 ) = 0;  = 1319.294, ω = 0.10142578, β = 0.0 0 01, η =
0.62811041, β1 = 0.0 0 01, k1 = 0.07142238, k2 = 0.014285824, γ , =
We now take into accounts the existence of the adjoint function 0.0374, v1 = 0.20 0 05051, v2 = 0.79999398, φ = 0.1050 0 0 05, =
λi , i = 1, 2, 3 . . . 6, such that they satisfy the equations; 0.0805840, α = 0.0 0440 0 0 0, m1 = 0.7178040 0, m2 =
0.52152716, τ1 = 0.290 0 0 0 0 for the numerical simulations of
dλ1 ∂H the optimal control problem. From Figs. 9a–9 d, the solution
= − ,
dt ∂S trajectories which considers the individual effects of each of the
= λ1 [ω + (1 − u2 )β (I + ηA ) + (1 − u2 − u3 )β1V  ] control variables, ui , i = 1, 2, . . . , 5 on the E, A, I and V population
were figured out respectively. The red lines depict the uncontrolled
−λ2 [ω + (1 − u2 )β (I + ηA ) + (1 − u2 − u3 )β1V  ];
population while the blue lines show the controlled population.
dλ2 ∂H These control measures were all highly significant in reducing the
= − ,
dt ∂E asymptomatic and the virus in the populations as compared to the
= −1 + λ2 [k1 γ + k2 (1 − γ ) + ω] − λ3 k2 (1 − γ ) − λ4 k1 γ ; exposed and severely infected populations. It can be inferred here
that, adapting to these measures effectively minimize the rate of
dλ3 ∂H
= − , virus into the environment decreasing the rate of transmission
dt ∂A through the environment which was very sensitive in the previous
= −1 + λ1 (1 − u2 )βηS − λ2 (1 − u2 )βηS analysis, see Fig. 4. More to this, Fig. 10 demonstrates the solution
+λ3 [ω + v1 φ + v2 (1 − φ )] − λ4 v1 φ trajectories by implementing all the control variables on the E, A,
−λ5 v2 (1 − φ ) − λ6 (1 − u4 )m1 ; I and V populations with increase in time. Similarly, the red line
represents the uncontrolled case while the blue line illustrates the
dλ4 ∂H collective controlled state.
= − ,
dt ∂I Figs. 9e, f, 10e and f illustrate the optimal profile effects of re-
= −1 + λ1 (1 − u2 )β S − λ2 (1 − u2 )β S spective controls u1 , u2 , u3 , u4 and u5 on COVID-19 transmission
+λ4 [ + ω + α ] − λ5 − λ6 (1 − u4 )m2 ; dynamics in the population and the environment. Fig. 11 repre-
sents the effect of control u1 on COVID-19 transmission dynam-
dλ5 ∂H
= − , ics in the population and the environment. Comparisons of the
dt ∂R dynamics of the subpopulations of exposed, asymptomatic, and
= λ5 ω; symptomatic individuals and the number of virus with control
dλ6 ∂H u2 and without any control implementation is demonstrated by
= − ,
dt ∂V Fig. 12. Figs. 13–15 shows the respective dynamics when control
= −1 + λ1 (1 − u2 − u3 )β1 S − λ2 (1 − u2 − u3 )β1 S u3 , u4 and u5 is used separately. It is revealed that the applica-
tion of the control delayed the COVID-19 virus to peak, and sig-
+λ6 (u5 + τ1 );
nificantly reduces the virus prevalence in the population and the
size of the virus present in the community. It is observed that the
with the transversality condition λi (t f ) = 0 given that ∀ui where
numbers of exposed, asymptomatic, severely infected individuals,
i = 1, 2, . . . , 5, we have
as well as the number of virus with control diminish more rapidly
∂H than when there is no control. It is seen that the separate use of
= 0. any of the three controls causes a sharp decrease in the number
∂ ui
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 13

Fig. 9. Solution trajectories showing the effects of the combined efforts of optimal control strategies ui (i = 1, 2, . . . , 5 ) on the E, A, I and V populations for 90 days.

of human infections as well as the number of virus presents in In a nut shell, the results of our optimal control analysis guaran-
the community. To achieve an optimal control of COVID-19 at a tee that, ensuring the effective utilization of the above-mentioned
minimized cost, then the following ideas deduced from numeri- optimal control strategies will significantly contribute to the reduc-
cal simulation’s results in Figs. 9e, f, 10 e and f can be adhered tion in the spread within the susceptible population. This will far
to: Considering the trajectory of the control measure u1 , it can be help to minimize the numerous effects which might have been ex-
seen that, u1 should be maintained at 0.5005 for the whole pe- perienced in the days ahead as far as the pandemic is concerned.
riod until the 90th day then maintained at 0.499 after day 90 as
shown in Fig. 10e. For the control measure u2 , efforts must be im-
4.2. Cost-effectiveness analysis
plemented around 0.5005. Similarly, efforts on u3 also behave like
u2 . The major differences are that; the maximum effort which can
Here, cost-effectiveness analysis is carried out based on the
be applied on u3 is around 0.50046 as compared to 0.5005 in u2 as
numerical implementation of the optimality system conducted in
shown in Fig. 10e and f. Given that the disease continues to per-
Section 4. The cost benefits associated with the implementation of
sist within the 90 days, then the regulation on u4 is that, efforts
the control strategies can be compared through cost-effectiveness
must start from 0.4996 after day 90 as shown in Fig. 10f. Lastly,
analysis. Thus, following the approach used in several previous
efforts on u5 must start around 0.4998 and gradually increased to
studies [53–56], the incremental cost-effectiveness ratio (ICER) is
around 0.5008. This effort has to be also maintained until the 90th
calculated to determine the most cost-effective strategy of all the
day and maintained at 0.4998 after day 90 as shown in Fig. 10f.
different control intervention strategies considered in this work.
14 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Fig. 10. Solution trajectories showing the effects of the combined efforts of optimal control strategies ui (i = 1, 2, . . . , 5 ) on the E, A, I and V populations for 300 days.

Most often, ICER is employed to measure up the changes between Table 4

Strategies 1–6 in increasing order of the number of infections averted.
the costs and the health benefits of any two different control inter-
vention strategies i and j competing for the same limited resources. Strategy Total infections averted Total cost ($) ICER
ICER is defined mathematically as Strategy 6 4.4136 × 109
3.35 7.5902 × 10−10
Strategy 4 7.0065 × 109 0.05 −1.2727 × 10−9
Difference in costs of control strategies i and j Strategy 3 7.0550 × 109 0.80 1.5464 × 10−8
ICER = .
Difference in infections averted by control strategies i and j Strategy 5 1.0748 × 1010 0.45 9.4774 × 10−11
Strategy 1 1.1681 × 1010 1.25 8.5745 × 10−10
(21)
Strategy 2 1.7117 × 1010 0.80 −8.2781 × 10−11
The numerator of ICER in Eq. (21) consists of the differences
in intervention costs, averted disease costs and costs of prevented only) and Strategy 6 (which combines the use of controls ui, i = 1,
cases among others. While the ICER denominator measures the dif- 2, … , 5) are ranked in increasing order of total number of infec-
ference in health outcome, which includes the total population of tions averted as presented in Table 4.
infection averted or total number of cases of susceptibility pre- Now, ICER is calculated for the intervention Strategies 6 and 4
vented as the case may be. in order to compare the two competing strategies incrementally as
Owing to the simulation results of the optimality system, the follows:
six control strategies, namely, Strategy 1 (u1 only), Strategy 2 (u2 3.35
only), Strategy 3 (u3 only), Strategy 4 (u4 only), Strategy 5 (u5 ICER(6 ) = = 7.5902 × 10−10 ,
4.4136 × 109
J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 15

Fig. 11. The optimal effect of applying only control u1 .

Fig. 12. The optimal effect of applying only control u2 .

16 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

Fig. 13. The optimal effect of applying only control u3 .

Fig. 14. The optimal effect of applying only control u4 .

 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 17

Fig. 15. The optimal effect of applying only control u5 .

Table 5 Table 6
Comparison between control intervention Strategies 4 and 3. Comparison between control intervention Strategies 4 and 5.

Strategy Total infections averted Total cost ($) ICER Strategy Total infections averted Total cost ($) ICER
−12
Strategy 4 7.0065 × 109
0.05 7.1362 × 10 Strategy 4 7.0065 × 109
0.05 7.1362 × 10−12
Strategy 3 7.0550 × 109 0.80 1.5464 × 10−8 Strategy 5 1.0748 × 1010 0.45 1.0691 × 10−10
Strategy 5 1.0748 × 1010 0.45 9.4774 × 10−11 Strategy 1 1.1681 × 1010 1.25 8.5745 × 10−10
Strategy 1 1.1681 × 1010 1.25 8.5745 × 10−10 Strategy 2 1.7117 × 1010 0.80 −8.2781 × 10−11
Strategy 2 1.7117 × 1010 0.80 −8.2781 × 10−11

Table 7
Comparison between control intervention Strategies 4 and 1.
0.05 − 3.35
ICER(4 ) = = −1.2727 × 10−9 , Strategy Total infections averted Total cost ($) ICER
7.0065 × 109 − 4.4136 × 109
0.80 − 0.05 Strategy 4 7.0065 × 10 9
0.05 7.1362 × 10−12
ICER(3 ) = = 1.5464 × 10−8 , Strategy 1 1.1681 × 1010 1.25 2.5671 × 10−10
7.0550 × 109 − 7.0065 × 109 Strategy 2 1.7117 × 1010 0.80 −8.2781 × 10−11
0.45 − 0.80
ICER(5 ) = = −9.4774 × 10−11 ,
1.0748 × 1010 − 7.0550 × 109 Table 8
1.25 − 0.45 Comparison between control intervention Strategies 4 and 2.
ICER(1 ) = = 8.5745 × 10−10 ,
1.1681 × 1010 − 1.0748 × 1010 Strategy Total infections averted Total cost ($) ICER
0.80 − 1.25
ICER(2 ) = = −8.2781 × 10−11 . Strategy 4 7.0065 × 10 9
0.05 7.1362 × 10−12
1.7117 × 1010 − 1.1681 × 1010 Strategy 2 1.7117 × 1010 0.80 7.4180 × 10−11

From Table 4, it is observed that the value of ICER(6) is greater

than that of ICER(4). This indicates that Strategy 6 is more costly interventions and ICER is recalculated for Strategies 4 and 1 as
and less effective than Strategy 4. For this reason, Strategy 6 is ex- shown in Table 7.
cluded from the list of alternative control interventions competing It can be seen in Table 7 that ICER(1) is greater than ICER(4).
for the same limited resources and ICER is recalculated for Strate- This implies that the implementation of Strategy 1 is more costly
gies 4 and 3 as illustrated by Table 5. and less effective than the implementation of Strategy 4. Hence,
It is seen from Table 5 that the value of ICER(3) is more than Strategy 1 is discarded from the list of alternative control inter-
the ICER(4) value. The implication of this is that Strategy 3 is vention strategies competing for the same limited resources. Now,
strongly dominated, more costly and less effective when compared the ICER is finally recalculated for Strategies 4 and 2 as shown in
with Strategy 4. Therefore, Strategy 3 is excluded from the list of Table 8.
alternative interventions and ICER is recalculated for Strategies 4 Table 8 reveals that ICER(2) is greater than ICER(4). Hence,
and 5. The results of the new computation is presented by Table 6. Strategy 2 is considered to be strongly dominated, more costly and
Table 6 shows that the ICER(4) is lower than ICER(5). Then, it less effective than Strategy 4. Consequently, Strategy 4 (cleaning
follows that Strategy 5 is more costly and less effective than Strat- of surfaces with home based detergents) is the most cost-effective
egy 4. Hence, Strategy 5 is excluded from the list of alternative strategy among all the six control intervention strategies under
consideration in this work.
18 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103

5. Concluding remarks Data availability statement

With reference to the recent travel restrictions in Ghana and The parameter values (data) used to support the findings of this
other parts of the world, we formulated COVID-19 dynamics in study have been described in Section 3 and Appendix A.
Ghana as a deterministic model. The purpose of the study was to
consider humans as key players in the transmission of this virus. Disclosure
The disease-free equilibrium and endemic equilibrium points were
found to be globally asymptotically stable. We found that, the ma- The authors fully acknowledge that this paper was developed
jor transmission parameters β , β 1 , m1 and m2 contributing to the as a result of the first and second author’s thesis and project work.
basic reproduction number of 1.99 − 3.37 were all attributed to
humans through personal contact with the susceptible class or ac- CRediT authorship contribution statement
tivities with the environs. It is further inferred from this study
that; applying optimal control strategy on the rate at which the Joshua Kiddy K. Asamoah: Conceptualization, Supervision, In-
virus is released into the system, m1 and m2 , and also on the rel- vestigation, Cost-effectiveness analysis, Writing - review & editing.
ative transmission rate due to human behavior will considerably Mark A. Owusu: Investigation, Validation. Zhen Jin: Funding ac-
strike down COVID-19 pandemic. quisition, Supervision, Review & editing. F.T. Oduro: Supervision.
It was also found that, it might be possible the recovered in- Afeez Abidemi: Cost-effectiveness analysis. Esther Opoku Gyasi:
dividuals can be reinfected, see Fig. 8b. When this happens, then Review & editing.
the number of the infected individuals will also increase. There-
fore, we highly recommend that, drug manufacturers should aim Declaration of Competing Interest
at drug samples which will induce permanent immunity in the
recovered individuals so as to reduce the susceptible population. The authors declare that they have no known competing finan-
Cost-effectiveness analysis was carried out based on the numerical cial interests or personal relationships that could have appeared to
implementation of the optimality system conducted in Section 4. influence the work reported in this paper.
This showed that, safety adopted by the asymptomatic and symp-
tomatic individuals such as practicing proper cough etiquette by Acknowledgment
maintaining a distance, covering coughs and sneezes with dispos-
able tissues or clothing and washing of hands after coughing or This research is supported by the African Institute for
sneezing is the most cost-effective strategy, followed by intensi- Mathematical Sciences. The National Natural Science Founda-
fying the usage of nose mask and face shields through educa- tion of China (11331009 and 61873154), Shanxi Key Laboratory
tion, then the effective testing and quarantine when boarders are (201705D111006), and Shanxi Scientific and Technology Innovation
opened, fumigating the commercial areas such as markets, clean- Team (201705D15111172). The first author is grateful to the Chinese
ing of surfaces with home-based detergents, and lastly, the combi- Government and the Complex Systems Research Center, Shanxi
nation of all the control interventions analysed in this study. It is University, for their support. The second author thank the African
highly guaranteed that, this study will help policy makers in the Institute for Mathematical Sciences for the support given, during
control of the pandemic in Ghana. Further research on this subject his postgraduate studies in 2020. The first author is grateful to
and other epidemiological study can be investigated, such as Baba Seidu, for his help during the review phase of the manuscript.

1. The impact of religious activities on the spread and control of

Appendix A. Confirmed cases for Ghana as of 7th May 2020.
the disease;
2. Optimal control model which seeks to reduce the disease in-
duced death rate.

Table A.9
Cumulative daily confirmed cases of COVID-19 as of 7th May, 2020 in Ghana.
Day 12th Mar 13th Mar 14th Mar 15th Mar 16th Mar 17th Mar 18th Mar

Cases 2 2 2 2 6 6 6
% Change - 0 0 0 200 0 0
Day 19th Mar 20th Mar 21st Mar 22nd Mar 23rd Mar 24th Mar 25th Mar
Cases 11 16 21 24 27 51 66
% Change 83.3 45.5 31.3 14.3 12.5 88.9 29.4
Day 26th Mar 27th 28th Mar 29th Mar 30th Mar 31st Mar 1st April
Cases 68 132 137 137 152 161 195
% Change 3.0 94.0 3.8 0 10.9 5.9 21.1
Day 2nd April 3rd April 4th April 5th April 6th April 7th April 8th April
Cases 204 205 214 214 287 313 313
% Change 4.6 0.5 4.4 0 34.1 9.1 0
Day 9th April 10th April 11th April 12th April 13th April 14th April 15th April
Cases 378 408 566 566 566 566 641
% Change 20.7 7.9 38.7 0 0 0 13.3
Day 16th April 17th April 18th April 19th April 20th April 21st April 22nd April
Cases 641 641 834 1042 1042 1042 1279
% Change 0 0 30.1 17.0 0 0 22.7
Day 23rd April 24th April 25th April 26th April 27th April 28th April 29th April
Cases 1279 1279 1550 1550 1671 2074 2074
% Change 0 0 21.2 0 7.8 24.1 0
Day 30th April 1st May 2nd May 3rd May 4th May 5th May 6th May
Cases 2074 2169 2719 2719 3091 3091 3091
% Change 0 4.6 25.4 0 12.0 0 0
Day 7th May
Cases 4012
% Change 12.0
 J.K.K. Asamoah, M.A. Owusu and Z. Jin et al. / Chaos, Solitons and Fractals 140 (2020) 110103 19

Supplementary material [28] Mandal M, Jana S, Nandi SK, Khatua A, Adak S, Kar T. A model based study on
the dynamics of covid-19: prediction and control. Chaos, Solitons & Fractals
2020:109889.
Supplementary material associated with this article can be [29] Abbasi Z, Zamani I, Mehra AHA, Shafieirad M, Ibeas A. Optimal control design
found, in the online version, at doi:10.1016/j.chaos.2020.110103. of impulsive SQEIAR epidemic models with application to COVID-19. Chaos,
Solitons & Fractals 2020:110054. doi:10.1016/j.chaos.2020.110054.
References [30] Yousefpour A, Jahanshahi H, Bekiros S. Optimal policies for control of the
novel coronavirus COVID-19 outbreak. Chaos, Solitons & Fractals 2020:109883.
doi:10.1016/j.chaos.2020.109883.
[1] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-1, 20
[31] Mandal M, Jana S, Nandi SK, Khatua A, Adak S, Kar T. A model based study on
january 2020. Geneva, Switzerland 2020.
the dynamics of COVID-19: prediction and control. Chaos, Solitons & Fractals
[2] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-3, 23
2020;136:109889. doi:10.1016/j.chaos.2020.109889.
january 2020. Geneva, Switzerland 2020.
[32] Ullah S, Khan MA. Modeling the impact of non-pharmaceutical inter-
[3] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-5, 25
ventions on the dynamics of novel coronavirus with optimal control
january 2020. Geneva, Switzerland 2020.
analysis with a case study. Chaos, Solitons & Fractals 2020:110075.
[4] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-8, 28
doi:10.1016/j.chaos.2020.110075. https://www.sciencedirect.com/science/
january 2020. Geneva, Switzerland 2020.
article/pii/S0960077920304720%2Fj.chaos.2020.109889
[5] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-11, 31
[33] Asamoah JKK, Jin Z, Sun G-Q, Li MY. A deterministic model for Q Fever trans-
january 2020. Geneva, Switzerland 2020.
mission dynamics within dairy cattle herds: using sensitivity analysis and op-
[6] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-12, 1
timal controls. Comput Math Methods Med 2020;2020:1–18. doi:10.1155/2020/
february 2020. Geneva, Switzerland 2020.
6820608.
[7] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-14, 3
[34] Korobeinikov A., Rezounenko A. Stability of a retrovirus dymanic model. arXiv
february 2020. Geneva, Switzerland 2020.
preprint arXiv:1812.11456; 2018.
[8] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-26,
[35] Smith HL, Thieme HR. Dynamical systems and population persistence, 118.
15th february 2020. Geneva, Switzerland 2020.
American Mathematical Soc; 2011.
[9] Worldometer. Covid-19 coronavirus pandemic. 2020. https://www.
[36] Jacquez JA, Simon CP. Qualitative theory of compartmental systems. SIAM Rev
worldometers.info/coronavirus/, Accessed 7th June 2020.
1993;35(1):43–79.
[10] Organization WH, et al. Novel coronavirus (2019-ncov) situation report-54, 14
[37] Atangana A, Igretaraz S. Mathematical model of COVID-19 spread in Turkey
march 2020. Geneva, Switzerland 2020.
and South Africa: Theory, methods and applications. medRxiv 2020. doi:10.
[11] Ghana Health Service. Covid-19 Ghana’s outbreak response management up-
1101/2020.05.08.20095588.
dates. https://ghanahealthservice.org/covid19/archive.php. Accessed 7th May
[38] Elazzouzi A, Alaoui AL, Tilioua M, Torres DFM. Analysis of a SIRI epidemic
2020.
model with distributed delay and relapse. Stat. Optim. Inform. Comput.
[12] BBC. Schools close down and social gathering restrictions. Retrieved from
2019;7(3). doi:10.19139/soic- 2310- 5070- 831.
https://www.bbc.com/pidgin/tori-51904164; Accessed 3rd May 2020.
[39] Son H. Analysis and optimal control of deterministic vector-borne diseases
[13] Kucharski AJ, Russell TW, Diamond C, Liu Y, Edmunds J, Funk S, et al. Early
model. 2018.
dynamics of transmission and control of COVID-19: a mathematical mod-
[40] Sene N. SIR Epidemic model with mittag–leffler fractional derivative. Chaos,
elling study. Lancet Infect Disea 2020;20(5):553–8. doi:10.1016/s1473-3099(20)
Solitons & Fractals 2020;137:109833. doi:10.1016/j.chaos.2020.109833.
30144-4.
[41] Martcheva M. An introduction to mathematical epidemiology, 61. Springer;
[14] Liu Y, Gayle AA, Wilder-Smith A, Rocklöv J. The reproductive number of
2015.
COVID-19 is higher compared to SARS coronavirus. J Travel Med 2020;27(2).
[42] Sene N. Stability analysis of the generalized fractional differential equations
doi:10.1093/jtm/taaa021.
with and without exogenous inputs. J Nonlinear Sci Appl 2019;12:562–72.
[15] Ivorra B, Ferrández M, Vela-Pérez M, Ramos A. Mathematical modeling of the
doi:10.22436/jnsa.012.09.01.
spread of the coronavirus disease 2019 (COVID-19) taking into account the un-
[43] Chibaya S, Nyabadza F. Mathematical modelling of the potential role of sup-
detected infections. the case of china. Commun Nonlinear Sci Numer Simul
plementary feeding for people living with hiv/aids. Int J Appl Comput Math
2020;88:105303. doi:10.1016/j.cnsns.2020.105303.
2019;5(3):97.
[16] Okuonghae D, Omame A. Analysis of a mathematical model for COVID-19 pop-
[44] Chibaya S, Nyabadza F. Mathematical modelling of the potential role of sup-
ulation dynamics in lagos, nigeria. Chaos, Solitons & Fractals 2020:110032.
plementary feeding for people living with HIV/AIDS. Int J Appl Comput Math
doi:10.1016/j.chaos.2020.110032.
2019;5(3):97. doi:10.1007/s40819- 019- 0660-9.
[17] Wang H, Wang Z, Dong Y, Chang R, Xu C, Yu X, et al. Phase-adjusted estimation
[45] World Population Review. Total population. https://worldpopulationreview.
of the number of coronavirus disease 2019 cases in wuhan, china. Cell Discov
com/. Accessed 5th May 2020.
2020;6(1). doi:10.1038/s41421- 020- 0148- 0.
[46] Bank T.W. Life expectancy at birth, total (years)-Ghana. https://data.worldbank.
[18] Higazy M. Novel fractional order SIDARTHE mathematical model of the COVID-
org/indicator/SP.DYN.LE00.IN?locations=GH Accessed 5th May 2020.
19 pandemic. Chaos, Solitons & Fractals 2020:110 0 07. doi:10.1016/j.chaos.2020.
[47] Center H.H.P.C.R. How long can the coronavirus that causes COVID-
109967.
19 survive on surfaces?. Retrieved from https://www.health.harvard.edu/
[19] Zhao Z, Li X, Liu F, Zhu G, Ma C, Wang L. Prediction of the COVID-19 spread
diseases- and- conditions/covid- 19- basics; Accessed 5th May 2020.
in african countries and implications for prevention and control: a case study
[48] Asamoah JKK, Oduro FT, Bonyah E, Seidu B. Modelling of rabies transmission
in south africa, egypt, algeria, nigeria, senegal and kenya. Sci Total Environ
dynamics using optimal control analysis. J Appl Math 2017;2017:1–23. doi:10.
2020;729:138959. doi:10.1016/j.scitotenv.2020.138959.
1155/2017/2451237.
[20] Ngonghala CN, Iboi E, Eikenberry S, Scotch M, MacIntyre CR, Bonds MH,
[49] Lenhart S, Workman JT. Optimal control applied to biological models. Chapman
et al. Mathematical assessment of the impact of non-pharmaceutical interven-
and Hall/CRC; 2007.
tions on curtailing the 2019 novel coronavirus. Math Biosci 2020;325:108364.
[50] Piguillem F., Shi L. Optimal COVID-19 quarantine and testing policies. CEPR
doi:10.1016/j.mbs.2020.108364.
Discussion Paper No. DP14613, Available at SSRN: https://ssrn.com/abstract=
[21] Goufo EFD, Khan Y, Chaudhry QA. HIV And shifting epicenters for COVID-19,
35942432020.
an alert for some countries. Chaos, Solitons & Fractals 2020:110030.
[51] Moore S.E., Okyere E. Controlling the transmission dynamics of covid-19. arXiv
[22] Asamoah J., Jin Z., Seidu B., Oduro F., Sun G.-Q., Alzahrani F. Mathemati-
preprint: 20 040 0443. 2020.
cal modelling and sensitivity assessment of covid-19 outbreak for Ghana and
[52] Asamoah JKK, Nyabadza F, Seidu B, Chand M, Dutta H. Mathematical modelling
Egypt. Available at SSRN 3612877; 2020.
of bacterial meningitis transmission dynamics with control measures. Comput
[23] Khan MA, Atangana A. Modeling the dynamics of novel coronavirus (2019-
Math Methods Med 2018;2018:1–21. doi:10.1155/2018/2657461.
ncov) with fractional derivative. Alexand. Eng. J. 2020. doi:10.1016/j.aej.2020.
[53] Tilahun GT, Makinde OD, Malonza D. Co-dynamics of pneumonia and typhoid
02.033.
fever diseases with cost effective optimal control analysis. Appl Math Comput
[24] Kassa SM, Njagarah JB, Terefe YA. Analysis of the mitigation strategies for
2018;316:438–59.
COVID-19: from mathematical modelling perspective. Chaos, Solitons & Frac-
[54] Abidemi A, Aziz NAB. Optimal control strategies for dengue fever spread in
tals 2020:109968. doi:10.1016/j.chaos.2020.109968.
Johor, Malaysia. Comput Methods Programs Biomed 2020;196:105585.
[25] Zhang R., Li Y., Zhang A.L., Wang Y., Molina M.J. Identifying airborne trans-
[55] Oke S, Matadi M, Xulu S. Cost-effectiveness analysis of optimal control strate-
mission as the dominant route for the spread of COVID-19. Proceedings of the
gies for breast cancer treatment with ketogenic diet. Far East J Math Sci
National Academy of Sciences 2020.
2018;109(2):303–42.
[26] Ferretti L, Wymant C, Kendall M, Zhao L, Nurtay A, Abeler-Dörner L,
[56] Olaniyi S, Okosun K, Adesanya S, Lebelo R. Modelling malaria dynamics with
et al. Quantifying SARS-CoV-2 transmission suggests epidemic control with
partial immunity and protected travellers: optimal control and cost-effective-
digital contact tracing. Science 2020;368(6491).
ness analysis. J Biol Dyn 2020;14(1):90–115.
[27] Lalwani S, Sahni G, Mewara B, Kumar R. Predicting optimal lockdown period
with parametric approach using three-phase maturation sird model for COVID-
19 pandemic. Chaos, Solitons & Fractals 2020:109939. doi:10.1016/j.chaos.2020.
109939.