Applied Mathematics and Computation 316 (2018) 438–459

Contents lists available at ScienceDirect

Applied Mathematics and Computation

journal homepage: www.elsevier.com/locate/amc

Co-dynamics of Pneumonia and Typhoid fever diseases with

cost effective optimal control analysis
Getachew Teshome Tilahun a,∗, Oluwole Daniel Makinde b, David Malonza c
a
Pan African University Institute of Basic Sciences Technology and Innovation, Nairobi, Kenya
b
Faculty of Military Science, Stellenbosch University, South Africa
c
Department of Mathematics, Kenyatta University, Kenya

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

Keywords: In this paper, we propose and analyze a mathematical model for Pneumonia–Typhoid co-
Co-infection infection to examine their characteristic relationship due to preventive and treatments
Pneumonia
strategies. Firstly, we qualitatively analyzed the model, and the basic reproduction num-
Typhoid fever
ber is determined with respect to the existence and stability of equilibria. The possibil-
Optimal control
Cost effectiveness ity of bifurcation is studied and also we investigated the sensitivity index of co-infection
Numerical simulation model reproduction number together with the local and global stability of the equili-
bra. Secondly, we extend the co-infection model by incorporating time dependent con-
trols as intervention and we used Pontryagin’s maximum principle for derivation of nec-
essary conditions for the optimal control and optimality system. Finally, the optimality
system is numerically simulated by considering four strategies and also their cost effec-
tiveness is analysed. We found that Pneumonia treatment with Typhoid fever prevention
costs least. Therefore, for optimal cost effective control of both diseases, the policy mak-
ers must focus more on prevention strategy while treating the infected individual is not
neglected.
© 2017 Elsevier Inc. All rights reserved.

1. Introduction

Pneumonia is classified as an airborne disease and most of the time it is acquired via inhalation or aspiration of pul-
monary pathogenic organisms into a lung segments [1]. The disease can be characterized by an inflammatory condition of
one or both of lungs. It can occur in all stages of human being but it is the most dangerous disease in older adults, ba-
bies, and people with other diseases or impaired immune systems [2]. Mainly Pneumonia is caused by bacteria and viruses.
Fungi and parasites are also responsible for the causes of this disease. One of the most responsible cause of Pneumonia is
calledStreptococcus Pneumoniae, it is bacterial pneumonia.
Typhoid fever is a bacterial infection that can spread throughout the body affecting many organs. Without prompt treat-
ment it can cause serious complications and can be fatal [3]. Typhoid fever is caused by virulent bacteria called Salmonella
typhi (S. typhi).

∗
Corresponding author.
E-mail address: [email protected] (G.T. Tilahun).

http://dx.doi.org/10.1016/j.amc.2017.07.063
0 096-30 03/© 2017 Elsevier Inc. All rights reserved.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 439

Mathematical modelling plays an important role in increasing our understanding of the dynamics of infectious diseases
and also to investigate the optimal use of intervention strategies to control the spread of infectious diseases. A lot of schol-
ars developed a mathematical model to describe the dynamics of the disease that helped them to propose disease control
mechanism and also described the dynamics of the co-infection with other infectious diseases. Some of them are [4–7] pro-
posed mathematical models that played significant roles in predicting suitable control strategies and analysing and ranking
their cost-effectiveness strategies in mitigating the disease. Also, Akinyi et al. [8] developed a mathematical model for the
impact of misdiagnosis and treatment of pneumonia as malaria. The result of the study showed that accurate diagnosis
of both pneumonia and malaria can be reduced if the basic reproduction number is reduced below unity. David Onyinge
and Omolo Ongati [6] proposed a mathematical model for the co-infection of Pneumonia–HIV/AIDS. Other studies [9], the
authors developed a deterministic mathematical model for assessing the transmission dynamics of typhoid in malaria en-
demic settings. The result of the study suggests that a Typhoid fever outbreak in malaria endemic settings may lead to
higher population of dually infected individuals displaying clinical symptoms of both infections than the singly-infected
population displaying clinical symptoms of the diseases. Adeboye and Haruna [10] developed SIRS mathematical model that
addressed the control of the transmission of Typhoid fever and Malaria simultaneously and also Akinyi et al. [11] propose a
mathematical model for Malaria–Typhoid co-infection.
However, all of them did not consider optimal control strategies in their studies, and also there are few mathematical
studies have been undertaken with co-infection of pneumonia with Malaria and also co-infection of Typhoid fever with
Malaria. But still no work has been undertaken to describe co-dynamics of Pneumonia and Typhoid fever diseases with cost
effective optimal control analysis to the best of our knowledge .
In this paper, an SIR model is formulated for Pneumonia–Typhoid fever co-infection with optimal problem and cost-
effectiveness analysis of the optimal intervention strategies are presented. This paper is arranged as follows: in Section 2,
we derive a model consisting of ordinary differential equations that describes the interactions and the dynamics of the
diseases with the underlying assumptions. Section 3 is devoted to qualitative analysis of the model and sensitivity analysis
of parameters. In Section 4, we use Pontryagin’s s Maximum Principle to investigate optimal strategies and to find the
necessary conditions for the optimal control of the disease. In Section 5, we show the simulation results. In Section 6 cost
effectiveness analysis is described. Our conclusions are discussed in Section 7.

2. Model formulation

2.1. Description of the model

The model consider the Typhoid fever causing bacteria (salmonella) population (B) and the human population . The hu-
man population is divided in to seven classes, susceptible (S), Pneumonia infectious (Ip ), Typhoid infectious (It ), pneumonia
and Typhoid co-infectious (Itp ), Pneumonia recovered (Rp ), Typhoid recovered (Rt ) and Pneumonia–Typhoid co-infectious re-
covered (Rtp )is also considered. The recruitment rate susceptible individuals either by birth or immigration is π and the
number of susceptible increases by those individuals that lost their temporary immunity from recovered sub classes of
Pneumonia (Rp ), Typhoid fever (Rt ) and Pneumonia–Typhoid fever co-infected sub class (Rtp ) with a rate of δ 1 , δ 2 and δ 3
γ (I +ϒ I )
respectively. Any susceptible individuals either can get Pneumonia disease with force of infection λ p = p
N
tp
and join
Pneumonia infectious sub-class (Ip ) or Typhoid fever disease with force of infection λt = k+B and join Typhoid fever in-
v B

fectious sub-class(It ), where Y > 1 is a modification parameter for the dually infected due to increased chance to infect
susceptible than Pneumonia only infected. And also γ is infectious rate of Pneumonia, v is the rate of ingestion of Typhoid
causing bacteria, k is concentration of bacteria in foods and water. The number of co-infection sub- class is increased from
Pneumonia infected group by getting Typhoid fever disease with λt force of infection and also from Typhoid fever infected
sub-class by getting Pneumonia disease with λp force of infection. The infectious sub-class of Pneumonia also can get treat-
ment with β 1 rate and move to Pneumonia recovered sub-class Rp or dies due to disease causing death rate of α 1 . Similarly
the infected sub-class of Typhoid fever also can get treatment with a rate of β 2 and join Typhoid recovered sub-class or
dies from disease causing death with a rate of α 2 . The Pneumonia- Typhoid co-infected sub class can get treatment with a
rate of σ and get temporary immunity either from both disease or Pneumonia only or Typhoid only and join the co-infected
recovered sub-class (Rtp ) with probability of (1 − g)(1 − e ) or Pneumonia recovered sub-class (Rp ) with probability of e or
Typhoid fever recovered sub-class (Rt ) with probability of g(1 − e ), where (1 − g)(1 − e ) + g(1 − e ) + e = 1. Additionally, in-
dividuals in the co-infected sub class also dies either Pneumonia or Typhoid causing death with similar rate of Pneumonia
or Typhoid only infected individuals. In all the seven human population sub-classes μ is natural causing death rate. The Ty-
phoid fever causing bacteria (salmonella) population (B) grows in contaminated food or drinks with Q rate and also increase
its number from the discharge of bacteria from Typhoid fever infected individuals and the co-infected individuals with a
rate of σ 1 and σ 2 respectively and also it dies due to Natural/ drug induced death rate of μb .
The above description of the model is plotted in Fig. 1 and we generate the following model as seven system of differ-
ential equation.
440 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

Fig. 1. Flow diagram of the model.

⎧
⎪ dS
⎪ = π + δ1 R p + δ2 Rt + δ3 Rt p − (λ p + λt + μ )S
⎪
⎪
⎪ dt
⎪
⎪ dIp
⎪
⎪
⎪ dt = λ p S − (λt + β1 + α1 + μ )Ip
⎪
⎪
⎪
⎪
⎪ dIt
⎪
⎪
⎪ = λt S − (λ p + β2 + α2 + σ1 + μ )It
⎪
⎪ dt
⎪
⎪
⎪
⎪ dI
⎨ t p = λt Ip + λ p It − (σ + α1 + α2 + σ2 + μ )It p
dt
(1)
⎪
⎪ dR p
⎪
⎪ = β1 I p + σ eIt p − (δ1 + μ )R p
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt = β2 It + σ g(1 − e )It p − (δ2 + μ )Rt
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt p
= σ (1 − g)(1 − e )It p − (δ3 + μ )Rt p
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎩ dB = Q + σ I + σ I − μ B,
1 t 2 tp b
dt

3. Qualitative analysis

In this section, we present some basic qualitative properties of the model. These analysis seek to show that the model is
epidemiologically appropriate in the sense the model and its predictions make sense. These analysis include finding the set
inside which the model can be sufficiently studied (i.e., the invariant region); and local and global stability of critical points
of the model.

3.1. Invariant region

In this section, we obtain a region in which the solution of (1) is bounded. For this model the total human population
is N = S + I p + It + It p + R p + Rt + Rt p . Then, after differentiating N with respect to time and substituting the expression for
dS dI p dIt dIt p R p dRt dRt p
,
dt dt
, dt , dt
, dt , dt and dt
and from Eq. (1) we obtain;

dN
= π − μN − α1 (It + It p ) − α2 (It + It p ). (2)
dt
If there is no death from Pneumonia and Typhoid fever, Eq. (2) become;
dN
≤ π − μN. (3)
dt
After, solving Eq. (3) and evaluating it as time tends to infinity, we got;

h = {(S, I p , It , It p , R p , Rt , Rt p ) ∈ 7 : 0 ≤ N ≤ πμ }.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 441

Similarly for bacteria population, if there is no discharge of bacteria from the infectious, then dB
dt
= Q + σ1 It + σ2 It p − μb B
become;
dB
≤ Q − μb B (4)
dt
After, solving Eq. (4) and evaluating it as time tends to infinity, we got;

b = {B ∈  : 0 ≤ Q ≤ μQ }.
b
Therefore, all the solution set of (1) is bounded in D =
h ×
b . Thus, it is sufficient to study the dynamics of the
model in D, inside which the model is considered to be epidemiologically and mathematically well-posed.

3.2. Positivity of the solution

In this section, we show all the solution of the model (1) remain positive for future time if their respective initial values
are positive.

Theorem 1. If S0 > 0, I p0 > 0, It0 > 0, It p0 > 0, R p0 > 0, Rt0 > 0, Rt p0 > B0 > 0 then all the solution set (S(t), Ip (t), It (t), Itp (t),
Rp (t), Rt (t), Rtp (t), B(t)) are positive for future time.

Proof. To proof this theorem first let as define t1 , t1 = sup{t > 0 : S(τ ) > 0, I p (τ ) > 0, It (τ ) > 0, It p (τ ) > 0, R p (τ ) >
0, Rt (τ ) > 0, Rt p (τ ) > 0, for all τ ∈ [0, t]}.
Since S0 ≥ 0, I p0 ≥ 0, It0 ≥ 0, It p0 ≥ 0, R p0 ≥ 0, Rt0 ≥ 0, Rt p0 ≥ 0, B0 ≥ 0 , thus t1 > 0. If t1 < ∞, then necessarily S or Ip or It
or Itp or Rp or Rt or Rtp or B is equal to zero at t1 . From Eq. (1), let as take the first equation,

dS
= π + δ1 R p + δ2 Rt + δ3 Rt p − (λ p + λt + μ )S. (5)
dt
Using the variation of constants formula the solution of Eq. (5) at t1 is given by.
  t1
 t  t 
1 1
S(t1 ) = S(0 )exp − (λ p + λt + μ )(s )ds + ( pi + δ1 R p + δ2 Rt + δ3 Rt p )exp − (λ p + λt + μ )(τ )dτ ds.
0 0 s

Moreover, since all the variables are positive in [0, t1 ], then S(t1 ) > 0.
It can be shown in a similar way that Ip (t1 ) > 0, It (t1 ) > 0, Itp (t1 ) > 0, Rp (t1 ) > 0, Rt (t1 ) > 0, Rtp (t1 ) > 0 and B(t1 ) > 0 which
is a contradiction. Hence t1 = ∞.
Therefore, all the solution sets are positive for future time. 

3.3. Disease free equilibrium

The disease free equilibrium of Eq. (1) is obtained by equating all equations of the model to zero and then letting I p =
0, It = 0 and It p = 0. Then we obtain;
π

E0 = , 0, 0, 0, 0, 0, 0, 0 .
μ

3.4. Basic reproduction number (0 )

By the principle of next generation matrix [12], we consider the following equations from the model.
dI p
= λ p S − (λt + β1 + α1 + μ )Ip
dt
dIt
= λt S − (λ p + β2 + α2 + σ1 + μ )It
dt
dIt p
= λt Ip + λ p It − (σ + α1 + α2 + σ2 + μ )It p
dt
dB
= Q + σ1 It + σ2 It p − μb B
dt
Then the next generation matrices are given by:
⎛γπ γπ ⎞ ⎛ ⎞
μ 0 μ 0 (β1 + α1 + μ ) 0 0 0
vπ (β2 + α2 + σ1 + μ )
⎜0 0 0 μk ⎟&V = ⎜ 0 0 0 ⎟
F =⎝ ⎠ ⎝ ⎠
0 0 0 0 0 0 (σ + α1 + α2 + σ2 + μ ) 0
0 0 0 0 0 −σ1 −σ2 μb
442 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

⎛ γπ γπ ⎞
0 0
⎜ μ ( β + α1 + μ ) μ(σ + α1 + α2 + σ2 + μ )
vπ ⎟
1
−1 ⎜ σ1 v π σ2 v π ⎟
FV = ⎜ 0
⎝ μμb k(β2 + α2 + σ1 + μ ) μμb k(β2 + α2 + σ1 + μ ) μμb k ⎟
⎠
0 0 0 0
0 0 0 0
Then the eigenvalues of F V −1 are obtained and given below;
γπ
λ∗1 = = 0 p
μ(β1 + α1 + μ )
σ1 v π
λ∗2 = = 0t
μμb k(β2 + α2 + σ1 + μ )
λ3,4 = 0
∗

According to the principle of next generation matrix, the basic reproduction number is the spectral radius of F V −1 , which
is :
0 = max{0 p , 0t }.

3.5. Local stability of disease free equilibrium

Theorem 2. The disease free equilibrium point is locally asymptotically stable if 0 < 1 and unstable if 0 > 1.
Proof. To proof this theorem first we obtained the Jacobian matrix of the model at the disease free equilibrium (E0 ) and is
given below in Eq. (6);
⎡−μ −γ π
0 −γ π
δ1 δ2 δ3 −vπ ⎤
μ μ μk
−γ π −γ Q
⎢ 0 μ − ( β1 + α 1 + μ ) 0 μ 0 0 0 0 ⎥
⎢ vπ⎥
⎢ 0 0 − ( β2 + α 2 + σ 1 + μ ) 0 0 0 0 μk⎥
⎢ ⎥
JE0 = ⎢ 0 0 0 − ( σ + α 1 + α 2 + σ2 + μ ) 0 0 0 0 ⎥ (6)
⎢ β1 − ( δ1 + μ ) 0 ⎥
⎢ 0 0 c1 0 0 ⎥
⎢ 0 0 β2 c2 0 − ( δ2 + μ ) 0 0 ⎥
⎣ ⎦
0 0 0 c3 0 0 − ( δ3 + μ ) 0
0 0 σ1 σ2 0 0 0 μb
The characteristic polynomial of Eq. (6) is;
γπ
 
(−μ − λ∗ )(−c1 − λ∗ ) ( − c3 ) − λ∗ (−c2 − λ∗ ) (λ∗ )3 + k1 (λ∗ )2 + (k2 + k3 (1 − 0t ))λ∗ − k4 (1 + 0t ) = 0,
μ
(7)
where,
c1 = δ3 + μ,
c2 = δ2 + μ,
c3 = β1 + α1 + μ,
k1 = β2 + 2α2 + σ1 + 2μ + σ + α1 + σ2 + μb ,
k2 = (σ + α1 + α2 + σ2 + μ )(β2 + α2 + σ1 + 2μ ),
k3 = μb (β2 + α2 + σ1 + μ ),
k4 = (μb (β2 + α2 + σ1 + μ )(σ + α1 + α2 + σ2 + μ ).
From Eq. (7) it is clear that;
−μ − λ = 0, ⇒ λ∗1 = −μ < 0
or
−c1 − λ = 0, ⇒ λ∗2 = −c1 < 0
or
−γ Q

− c3 − λ∗ = 0, ⇒ λ∗3 = c3 (0 p − 1 ), λ∗3 < 0 if 0 p < 1

μ
or
(λ∗ )3 + k1 (λ∗ )2 + (k2 + k3 (1 − 0t ))λ∗ − k4 (1 + 0t ) = 0. (8)
When we apply Routh–Hurwitz criteria on Eq. (8) it have strictly negative real root if 0t < 1.
Therefore the disease free equilibrium is locally asymptotically stable if 0p < 1 and 0t < 1 which means that 0 < 1. 
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 443

3.6. Global asymptotic stability of disease free equilibrium

To investigate the global stability of disease free equilibrium we used technique implemented by Castillo-Chavez and
Song [13]. First the full Pneumonia–Typhoid fever model (1) can be re-written as:

dX
= F (X, Z ),
dt

dZ
= G(X, Z ), G(X, 0 ) = 0.
dt
Where, X stands for the uninfected population, that is X = (S, R p , Rt , Rt p ) and Z also stands for the infected population ,that
is Z = (I p , It , It p ). The disease free equilibrium point of the model is denoted by U = (X ∗ , 0 ).
The point U = (X ∗ , 0 ) to be globally asymptotically stable equilibrium for the model provided that 0 < 1 (which is locally
asymptotically stable) and the following conditions must be met:
(H1 ) . For dX dt
= F (X, 0 ), X ∗ is globally asymptotically stable.
(H2 ) . G(X, Z ) = AZ − G˜ (X, Z ), G˜ (X, Z ) ≥ 0 for (X, Z) ∈

If the model (1) met the above two criteria then the following theorem holds:

Theorem 3. The point U = (X ∗ , 0 ) is globally asymptotically stable equilibrium provided that 0 < 1 and the condition (H1 ) and
(H2 ) are satisfied.

Proof. From system (1) we can get F(X, Z) and G(X, Z);
⎛ ⎞
π + δ1 R p + δ2 Rt + δ3 Rt p − (λ1 + λ2 + μ )S
⎜ ⎟
⎜ β1 Ip + σ eIt p − (δ1 + μ )R p ⎟
F (X, Z ) = ⎜
⎜
⎟
⎟
⎝ β2 It + σ g(1 − e )It p − (δ2 + μ )Rt ⎠
σ (1 − g)(1 − e )It p − (δ3 + μ )Rt p
⎛ ⎞
λ1 S − (λ2 + β1 + α1 + μ )Ip
⎜ ⎟
G(X, Z ) = ⎝ λ2 S − (λ1 + β2 + α2 + σ1 + μ )It ⎠
λ2 Ip + λ1 It − (σ + α1 + α2 + σ2 + μ )It p
Consider the reduced system,
⎛ ⎞
 π − μS
dX  ⎜ 0 ⎟
dt
 =⎝
0 ⎠
(9)
Z=0
0

From Eq. (9), it is obvious that X ∗ = ( π μ , 0 ) is the global asymptotic point. This can be verified from the solution, namely
S= πμ + ( S ( 0 ) − π )e−μt . As t → ∞, the solution (S) → π , implying that the global convergence of (9) in
.
μ μ
Let
 
γ − (β1 + α1 + μ ) 0 γ
A= 0 −(β2 + α2 + σ1 + μ ) 0
0 0 −(σ + α1 + α2 + σ2 + μ )

. Then, G(X, Z) can be written as, G(X, Z ) = AZ − G˜ (X, Z ), where,

˜   
G1 (X, Z ) γ (It + It p )(1 − NS ) + λ2 Ip
G˜ (X, Z ) = G˜2 (X, Z ) = λ1 It (10)
G˜3 (X, Z ) −(λ2 I p + λ1 It )

In Eq. (10) G˜2 (X, Z ) < 0 which leads to G˜ (X, Z ) < 0, that means the second condition (H2 ) is not satisfied, so U = (X ∗ , 0 ) may
not be globally asymptotically stable when 0 < 1. 

3.7. The endemic equilibrium

The endemic equilibrium is denoted by Et∗p = (S∗ , I∗p , It∗ , It∗p , R∗p , Rt∗ , Rt∗p , B∗ ) and it occur when the disease persist in the
community. To obtain it we equate all the model Eq. (1) to zero. Then we obtain:
444 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

π c(λ∗p + k2 )(λt∗ + k1 )k3

S∗ =
c (λ∗p + λt∗ + μ )(λ∗p + k2 )(λt∗ + k1 )k3 − [(λ∗p + k2 )(c1 k3 λ∗p + c3 λ∗p λt∗ ) + (λt∗ + k1 )(c2 k3 λt∗ + c3 λ∗p λ∗p )]

λ∗p π c(λ∗p + k2 )k3

I∗p =
c (λ + λ + μ )(λ + k2 )(λ + k1 )k3 − [(λ∗p + k2 )(c1 k3 λ∗p + c3 λ∗p λt∗ ) + (λt∗ + k1 )(c2 k3 λt∗ + c3 λ∗p λt∗ )]
∗
t
∗
p
∗
p
∗
p

λt∗ π c(λt∗ + k1 )k3

It∗ =
c (λ∗p + λ∗p + μ )(λ∗p + k2 )(λt∗ + k1 )k3 − [(λ∗p + k2 )(c1 k3 λ∗p + c3 λ∗p λt∗ ) + (λ∗p + k1 )(c2 k3 λt∗ + c3 λ∗1 λt∗ )]

λ∗p λt∗ π c(λ∗p + k2 )(λt∗ + k1 )

It∗p =
c (λ + λ + μ )(λ + k2 )(λ + k1 )k3 − [(λ∗p + k2 )(c1 k3 λ∗p + c3 λ∗p λt∗ ) + (λt∗ + k1 )(c2 k3 λt∗ + c3 λ∗p λt∗ )]
∗
p
∗
t
∗
p
∗
t

β1 I∗p + σ eIt∗p
R∗p =
δ1 + μ
β2 It∗ + σ g(1 − e )It∗p
Rt∗ =
δ2 + μ
σ (1 − g)(1 − e ) ∗
Rt∗p = It p
δ3 + μ
Q σ1 It∗ + σ2 It∗p
B∗ = +
μb μb
where,
k1 = β1 + α1 + μ

k2 = β2 + α2 + σ1 + μ

k3 = σ + α1 + α2 + σ2 + μ

c = (δ1 + μ )(δ2 + μ )(δ3 + μ )

c1 = δ1 β1

c2 = δ2 β2

c3 = δ1 σ e + δ2 σ g(1 − e ) + δ3 σ (1 − g)(1 − e )

3.8. Impact of Pneumonia on Typhoid fever infection

To describe impact of Pneumonia on Typhoid fever and vase versa, we express 0p interims of 0t .
Since ,
σ1 v π
0t =
μμb k(β2 + α2 + σ1 + μ )
σ1 v π
⇒μ= .
0t μb k(β2 + α2 + σ1 + μ )
Then substituting the expression for μ in 0p give
γ 0t μb k(β2 + α2 + σ1 + μ )
0 p = .
σ1 v(β1 + α1 + μ )
To investigate the impact of the two disease each other we did;
∂ 0 p γ μb k(β2 + α2 + σ1 + μ )
= > 0. (11)
∂ 0t σ1 v(β1 + α1 + μ )
Eq. (11) shows that Typhoid fever cases increase Pneumonia cases and also similarly Pneumonia cases increase Typhoid
fever cases.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 445

To investigate treatment of Pneumonia reduces Typhoid fever diseases;

From
γπ
0 p = ,
μ(β1 + α1 + μ )
we get
γπ
μ= . (12)
0 p (β1 + α1 + μ )
After combination of Eq. (12) and 0t we obtain,
σ1 v(0 p (β1 + α1 + μ ) + γ π )
0t = (13)
kμb (β1 + α1 + μ )(β1 + α1 )(β2 + α2 + μ + σ1 )

∂ 0t σ1 v
0 p (β 2 + 2β1 (α1 + μ ) + (α1 + μ )2 )) + γ π (2β1 + 2α1 + μ )
1
=− (14)
∂β1 kμb (β2 + α2 + μ + σ1 ) (β1 + α1 + μ )2 (β1 + α1 )2
∂
Eq. (14) shows ∂β0t < 0 , this mean treatment of Pneumonia (β 1 ) have an impact in decreasing Typhoid fever infection and
1
similarly can be shown treatment of Typhoid fever (β 2 ) have an impact in Pneumonia infection cases.

3.9. Bifurcation analysis

We investigate the nature of the bifurcation by using the method introduced in [13], which is based on the use of
the center manifold theory. In short, the method is summarized by Theorem 4.1 in [13] (for convenience, we recall it in
the appendix, see Theorem Appendix A.1). In such a theorem, there are two important quantities: the coefficients, say a
and b, of the normal form representing the dynamics of the system on the central manifold. These coefficients ‘decide’
the bifurcation. In particular, if a < 0 and b > 0, then the bifurcation is forward; if a > 0 and b > 0, then the bifurcation is
backward. Using this approach, the following result may be obtained.

Theorem 4. If 0 < 1 and

v∗ Q
a0 = σ2 μb β2 − σ1 (δ + μ ) > 0 (15)
μk 2
then system (1) exhibits a backward bifurcation at 0 = 1. If the inequality holds reversed, then the system exhibits a forward
bifurcation at 0 = 1.

Proof. First, we consider the transmission rate γ and v as bifurcation parameters so that 0 p = 1 and 0t = 1 if and only
if
μ(β1 + α1 + μ )
γ = γ∗ =
π
and
kμb (β1 + α1 + μ )(β2 + α2 + μ + σ1 )
v = v∗ = .
σ1 (0 p (β1 + α1 + μ )(β1 + α1 ) + γ π )
Then we make the following change of variables S = x1 , I p = x2 , It = x3 , It p = x4 , R p = x5 , Rt = x6 , Rt p = x7 , B = x8 . In addition,
using vector notation x = (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 )T , the Pneumonia Typhoid model can then be written in the form dx dt
=
F (x ), with F = ( f1 , f2 , f3 , f4 , f5 , f6 , f7 , f8 )T , as shown below:
⎧
⎪
⎪
dx1
= π + δ1 x5 + δ2 x6 + δ3 x7 − (λ px + λtx + μ )x1
⎪
⎪
⎪
⎪
dt
⎪ dx2
⎪
⎪
⎪ = λ px x1 − (λtx + β1 + α1 + μ )x2
⎪
⎪ dt
⎪
⎪ dx3
⎪
⎪ = λtx x1 − (λ px + β2 + α2 + σ1 + μ )x3
⎪
⎪ dt
⎪
⎪
⎪
⎨ dx4 = λtx x2 + λ px x3 − (σ + α1 + σ2 + α2 + μ )x4
dt (16)
⎪
⎪ dx5
= β1 x2 + σ ex4 − (δ1 + μ )x5
⎪
⎪
⎪ dt
⎪
⎪
⎪ dx6
⎪
⎪ = β2 x3 + σ g(1 − e )x4 − (δ2 + μ )x6
⎪
⎪ dt
⎪
⎪
⎪
⎪
dx7
= σ (1 − g)(1 − e )x4 − (δ3 + μ )x7
⎪
⎪
⎪
⎪
dt
⎪
⎩ dx8 = Q + σ 1 x 3 + σ 2 x 4 − μb x 8 ,
dt
446 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

vx8
where, λ px = γ (x2 + ϒ x4 ) and λtx = k+x8
.
This method involves evaluation of the Jacobian of system (16) at the disease free equilibrium (DFE), denoted by JE0 .
This becomes

⎡−μ −J1 0 −J2 δ1 δ2 δ3 −J3

⎤
⎢ 0 −J4 0 −J2 0 0 0 0 ⎥
⎢ 0 0 −J5 0 0 0 0 J6 ⎥
⎢ ⎥
⎢ 0 0 0 −J7 0 0 0 0 ⎥
JE0 = ⎢ ⎥ (17)
⎢ 0 β1 0 J11 −J8 0 0 0 ⎥
⎢ 0 0 β2 J9 0 −J9 0 0 ⎥
⎣ ⎦
0 0 0 J10 0 0 −J11 0
0 0 σ1 σ2 0 0 0 μb

where,
γπ γ ϒπ γπ
J1 = μ , J2 = μ , J3 = vμπk , J4 = μ + (β1 + α1 + μ ), J5 = (β2 + α2 + σ1 + μ ), J6 = vμπk ,
J7 = (σ + α1 + α2 + σ2 + μ ), J8 = (δ1 + μ ) J9 = (δ2 + μ ), J10 = β1 + α1 + μ and J11 = (δ3 + μ ). We first start by calculat-
ing the right eigenvector of J(E0 ) denoted respectively by w = [w1 , w2 , w3 , w4 , w5 , w6 , w7 , w8 , ]T .

⎡−μ −J1 0 −J2 δ1 δ2 δ3 −J3 w1

⎤⎛ 0
⎞ ⎛ ⎞
⎢ 0 −J4 0 −J2 0 0 0 0 ⎥⎜w2 ⎟ ⎜0⎟
⎢ 0 0 −J5 0 0 0 0 J6 ⎥⎜w3 ⎟ ⎜0⎟
⎢ ⎥⎜ ⎟ ⎜ ⎟
⎢ 0 0 0 −J7 0 0 0 0 ⎥⎜w4 ⎟ ⎜0⎟
⎢ β1 −J8
⎥⎜ ⎟ = ⎜ ⎟ (18)
⎢ 0 0 J11 0 0 0 ⎥⎜w5 ⎟ ⎜0⎟
⎢ 0 0 β2 J9 0 −J9 0 0 ⎥ ⎜ ⎟ ⎜ ⎟
⎣ ⎦⎝w6 ⎠ ⎝0⎠
0 0 0 J10 0 0 −J11 0 w7 0
0 0 σ1 σ2 0 0 0 −μb w8 0

Then Eq. (18) can be written as;

⎧
⎪−μw1 − J1 w2 − J2 w4 + δ1 ws + δ2 w6 + δ3 w7 − J3 w8 = 0
⎪
⎪
⎪
⎪−J4 w2 − J2 w4 = 0
⎪
⎪
⎪
⎪
⎪
⎪−J5 w3 − J6 w8 = 0
⎪
⎪
⎪
⎨−J7 w4 = 0
(19)
⎪
⎪β1 w2 + J11 w4 − J8 w5 = 0
⎪
⎪
⎪
⎪
⎪
⎪β2 w3 + J9 w4 − J9 w6 = 0
⎪
⎪
⎪
⎪J10 w4 − J11 w7 = 0
⎪
⎪
⎩
σ 1 w 3 + σ 2 w 4 − μb w 8 = 0 .

From Eq. (19) we obtain,

w1 = μ1
(δ2 w6 − J3 w8 ), w2 = w4 = w5 = w7 = 0, w3 = w3 > 0, w6 = βJ 2 w3 and w8 = μσ1 w3 .
9 b
The left eigenvectors of JE associated with the zero eigenvalue is given by v = (v1 , v2 , v3 , v4 , v5 , v7 , v8 )T , is calculated
as;

⎡−μ 0 0 0 0 0 0
⎤⎛v ⎞ ⎛0⎞
0 1
⎢ −J1 −J4 0 0 β1 0 0 ⎥⎜v2 ⎟ ⎜0⎟
0
⎢ 0 −J5 β2 σ1 ⎥⎜ ⎟ ⎜ ⎟
⎢ 0 0 0 0
⎥⎜v3 ⎟ ⎜0⎟
⎢ −J2 −J2 0 −J7 J11 J9 J10 σ2 ⎥⎜v4 ⎟ ⎜0⎟
⎢δ −J8
⎥⎜ ⎟ = ⎜ ⎟ (20)
⎢ 1 0 0 0 0 0 0 ⎥⎜v5 ⎟ ⎜0⎟
⎢ δ2 −J9 0 ⎥ ⎜ ⎟ ⎜ ⎟
⎣ 0 0 0 0 0 ⎦⎝v6 ⎠ ⎝0⎠
δ3 0 0 0 0 0 −J11 0 v7 0
−J3 0 J6 0 0 0 0 −μb v8 0
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 447

Eq. (20) can be written as,

⎧−μv = 0
⎪
⎪
1
⎪
⎪
⎪
⎪−J1 v1 − J2 v2 + β1 v5 = 0
⎪
⎪
⎪
⎪
⎪
⎪−J5 v3 + β2 v6 + σ1 v8 = 0
⎪
⎪
⎪
⎨−J2 v1 − J2 v2 − J7 v4 + J11 v5 + J9 v6 + J10 v7 + σ2 v8 = 0
(21)
⎪
⎪δ1 v1 − J8 v5 = 0
⎪
⎪
⎪
⎪
⎪
⎪δ2 v1 − J9 v6 = 0
⎪
⎪
⎪
⎪
⎪
⎪δ3 v1 − J11 v7 = 0
⎪
⎩
−J3 v1 + J6 v3 − μb v8 = 0.
σ2 J
Solving Eq. (21) gives, v1 = v2 = v5 = v6 = v7 = 0, v3 = v3 > 0, v4 = J7 v8 and v8 = σ1 v3 where
5
v3 is calculated to ensure
that the eigenvectors satisfy the condition v.w = 1.
The coefficients a and b are defined in Eqs. (22) and (23) below;


n
∂ 2 fk
a= vk wi w j ( S , 0, 0, 0, 0, 0, 0, 0 ), (22)
k, j,i=1
∂ xi ∂ x j 0


n
∂ 2 fk
b= vk wi ( S , 0, 0, 0, 0, 0, 0, 0 ), (23)
k,i=1
∂ xi ∂ v∗ 0

Taking into account of system (16) and considering only the non-zero components of the left eigenvector v, it follows that:
 
∂ 2 f3
a = v3 3w1 w8
∂ x1 ∂ x8
and
∂ 2 f3
b = v3 w8 .
∂ x8 ∂v∗
where the fi ’s denote the right hand side of system (16). It can be checked that:

∂ 2 f3 v
= ,
∂ x1 ∂ x8 k

∂ 2 f3
= .
∂ x 8 ∂ v ∗ μk
It follows:
σ1
b= vw
μμb k 3 3
so that b is always positive, and:
 
a0
a = 3 v ∗ σ1 w23 v3
μμ2b J9
where a0 is defined in (15). Therefore, system (1) exhibits backward or forward bifurcation at 0 = 1 according to the sign
of a0 . 

3.10. Sensitivity analysis

We did Sensitivity analysis of some basic parameters of the model. This helped us to identify the parameters that have
great influence on the basic reproductive number (0 ). We used the same technique outlined by Chitnis et al. [14]. Sensi-
 ∂
tivity index of 0 with respect to some parameter, say k is given by k 0 = ∂ k0 k . Since 0 =max{0p , 0t }, we obtain the
0
sensitivity analysis of 0p and 0t separately.
448 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

Table 1
Indices of sensitivity.

Symbol Description Sensitivity indices

0p Basic reproduction number of Pneumonia only

γ Infectious rate of Pneumonia 1
β1 Treatment rate of Pneumonia −036055
α1 Pneumonia induced death rate −0.91942
μ Natural causing death rate −1.044
0t Basic reproduction number of Typhoid fever only
v Ingestion rate of Salmonella bacteria 1
σ1 Discharge rate of Salmonella from Typhoid infected individuals 0.011
β2 Treatment rate of Typhoid fever −0.0025
α2 Typhoid fever induced death rate −0.00642
μb Natural/ drug induced death rate of Salmonella −1
μ Natural causing death rate −1.003

 ∂ 0 p γ π γ μ(β1 + α1 + μ )
γ 0 p = = =1>0
∂γ 0 p μ(β1 + α1 + μ ) γπ
 ∂ 0 p α1 α1
α10 p = =− <0
∂α1 0 p (β1 + α1 + μ )

 ∂ 0 p β1 β1
β10 p = =− <0
∂β1 0 p (β1 + α1 + μ )

 ∂ 0 p μ (β1 + α1 + 2μ )
μ0 p = =− <0
∂μ 0 p (β1 + α1 + μ )
∂ 0t v σ1 Q vμμb (β2 + α2 + σ1 + μ )
v 0t = = =1>0
∂v 0t μμb (β2 + α2 + σ1 + μ ) σ1 v π
∂ 0t σ1 (β2 + α2 + μ )
σ10t = = >0
∂σ1 0t (β2 + α2 + σ1 + μ )
∂ 0t α2 α2
α20t = =− <0
∂α2 0t (β2 + α2 + σ1 + μ )
∂ 0t β2 β2
β20t = =− <0
∂β2 0t (β2 + α2 + σ1 + μ )
∂ 0t μb σ1 v π μ2b μ(β2 + α2 + σ1 + μ )
μ0bt = =− 2 = −1 < 0
∂μb 0t μb μ(β2 + α2 + σ1 + μ ) σ1 v Q
∂ 0t μ (β2 + α2 + σ1 + 2μ )
μ0t = =− <0
∂μ 0t (β2 + α2 + σ1 + μ )
The above sensitivity analysis is summarized in Table 1.
In Table 1, parameters that have positive sensitivity indices, particularly γ and v have great influence in expanding the
disease if their values are increased, while the other parameters are unchanged. Due to the reason that, they increase the
average number of secondary case infection.
Examining the sensitivity analysis, it is not biologically reasonable to suggest that the human mortality rate (α 1 , α 2 , μ)
be increased in order to control the disease. The other possible sensitive parameters (negative sensitivity indices) that are
important for effective control of the disease are treatment rate of pneumonia, treatment rate of typhoid fever and drug
/natural induced death rate of salmonella. Therefore, in Section 4, we shall incorporate control measures to determine the
optimal strategy for controlling Pneumonia-Typhoid fever co-infection.

4. Extension of the model in to optimal control

In this section, we extend Pneumonia–Typhoid fever co-infection model (1) by incorporating five control interventions
. This helped us to identify the best intervention strategies that helps to eradicate the disease in the specified time. The
interventions are defined as:

i. u1 prevention effort of Pneumonia disease, that protect susceptible from contacting the disease.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 449

ii. u2 prevention of Typhoid fever disease by keeping proper personal sanitation that protect susceptible from contacting
Salmonella bacteria.
iii. u3 treatment effort of Pneumonia infected individuals.
iv. u4 treatment effort of typhoid fever infected individuals.
v. u5 mass cleaning of environments and ponds to eliminate Typhoid causing bacteria.

After incorporating, u1 , u2 , u3 , u4 and u5 in Pneumonia–Typhoid co-infection model (1), we obtain the following optimal
control model:

⎧
⎪ dS
⎪
⎪ = π + δ1 R p + δ2 Rt + δ3 Rt p − ((1 − u1 )λ p + (1 − u2 )λt + μ )S
⎪
⎪ dt
⎪
⎪
⎪
⎪ dI p
= (1 − u1 )λ p S − (1 − u2 )λt I p − (β1 + u3 )I p − (α1 + μ )I p
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dIt
= (1 − u2 )λt S − (1 − u1 )λ p I − t − (β2 + u4 )It − (α2 + σ1 + μ )It
⎪
⎪
⎪
⎪
dt
⎪
⎪
⎪ dI
⎨ t p = (1 − u2 )λt Ip + (1 − u1 )λ p It − (σ + u3 + u4 )It p − (α1 + α2 + σ2 + μ )It p
dt
(24)
⎪
⎪ dR p
⎪
⎪ = (β1 + u3 )I p + (σ e + u3 )It p − (δ1 + μ )R p
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt
⎪
⎪ = (β2 + u4 )It + (σ g(1 − e ) + u4 )It p − (δ2 + μ )Rt
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt p
= (σ (1 − g)(1 − e ) + u3 + u4 )It p − (δ3 + μ )Rt p
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎩ dB = Q + σ I + σ I − (u + μ )B,
1 t 2 tp 5 b
dt

To study the optimal levels of the controls the control set U is Lebesgue measurable and it is defined as:
U = {(u1 (t ), u2 (t ), u3 (t ), u4 (t ), u5 (t )) : 0 ≤ u1 < 1, 0 ≤ u2 < 1, 0 ≤ u3 < 1, 0 ≤ u4 < 1, 0 ≤ u5 < 1, 0 ≤ t ≤ T }. Our aim is to ob-
tain a control U and Ip , It and Itp that minimize the proposed J, and the form of the objective functional (J) is taken from
[15], given by:


1
tf 5
J= min (b1 Ip + b2 It + b3 It p + wi u2i )dt. (25)
u1 ,u 2 ,u 3 ,u 4 ,u 5 0 2
i=1

Where b1 , b2 , b3 and wi are positive. The expression 12 wi u2i represents cost which is associated with the controls ui . The
form is quadratic because we assume that costs are non-linear in its nature. Our aim is to minimize the number of infectious
and costs. Thus, we want to obtain an optimal controls (u∗1 , u∗2 , u∗3 , u∗4 , u∗5 ) in which:
J (u∗1 , u∗2 , u∗3 , u∗4 , u∗5 ) = min{J (u1 , u2 , u3 , u4 , u5 )/ui ∈ U }, where, U = {(u1 , u2 , u3 , u4 , u5 )/ each ui is measurable with
0 ≤ ui < 1 for 0 ≤ t ≤ tf .

4.1. The Hamiltonian and optimality system

By using Pontryagin’s Maximum Principle Pontryagin [16], we obtained a Hamiltonian (H) defined as:
dI dI dR dR
H (S, I p , It , It p , RP , Rt , Rt p , B, t ) = L(I p , It , It p , u1 , u2 , u3 , u4 , u5 , t ) + h1 ds
dt
+ h2 dtp + h3 dI
dt
t
+ h4 dtt p + h5 dtp + h6 dR
dt
t
+ h7 dtt p + h8 dB
dt
.
1 5
Where L(I p , It , It p , u1 , u2 , u3 , u4 , u5 , t ) = b1 I p + b2 It + b3 It p + 2 i=1 wi u2i , hi , i = 1, 2, 3, 4, 5, 6, 7, 8 are the adjoint variable
functions to be determined suitably by applying Pontryagin’s maximal principle [16] and also using [17] for existence of the
optimal control pairs.

Theorem 5. For an optimal control set u1 , u2 , u3 , u4 , u5 that minimizes J over U, there is an adjoint variables, h1 , . . . , h8 such
that:
450 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

⎧  
⎪
⎪ dh1 (1 − u2 )Bv γ (Ip + ϒ It p )(1 − u1 ) h2 γ (I p + ϒ It p )(1 − u1 ) h3 (1 − u2 )Bv
⎪ dt = −h1 − − −μ − −
⎪
⎪ k+B N N k+B
⎪
⎪

⎪
⎪ dh2 h1 γ (1 − u1 )S γ (1 − u1 )S (1 − u2 )Bv h3 γ (1 − u1 )It
⎪
⎪ = −b1 + − h2 − − β1 − u 3 − α1 − μ +
⎪
⎪ dt N N k+B N
⎪
⎪ 1 − u Bv γ 1 − u I

⎪
⎪ ( 2) ( 1) t
⎪
⎪ −h4 + − h 5 ( β1 + u 3 )
⎪
⎪ k+B N
⎪
⎪  
⎪
⎪ γ (Ip + ϒ It p )(1 − u1 ) h4 γ (I p + ϒ It p )(1 − u1 )
⎪
⎪ dh3
= −b2 − h3 − − β 2 − u 4 − α 2 − σ1 − μ − − h6 (β2 + u4 ) − h8 σ1
⎪
⎪
⎪
⎪ dt N N
⎪
⎪ γ 1 − u I

⎪
⎨ dh4 h1 γ (1 − u1 )S h2 γ (1 − u1 )S h3 γ (1 − u1 )It ( 1) t
= −b3 + − + − h4 − σ − u 3 − u 4 − α 1 − α 2 − σ2 − μ
dt N N N N (26)
⎪
⎪ −h5 (σ e + u3 ) − h6 (σ (1 − e ) + u4 ) − h7 (σ (1 − g)(1 − e ) + u3 + u4 ) − h8 σ2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ dh5
= −h1 δ1 − h5 (−δ1 − μ )
⎪
⎪
⎪
⎪
dt
⎪
⎪
⎪
⎪ dh6
= −h1 δ2 − h6 (−δ2 − μ )
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dh 7
= −h1 δ3 − h7 (−δ3 − μ )
⎪
⎪ dt
⎪
⎪
⎪
⎪
⎪
⎪ dh h1 (1 − u2 )Bv S h2 (1 − u2 )Bv I p h3 (1 − u2 )Bv S h4 (1 − u2 )Bv I p
⎩ 8 =− − + + − h8 (−u5 − μb )
dt (k + B )2 (k + B )2 (k + B )2 (k + B )2

With transversality conditions, hi (t f ) = 0, i = 1, . . . , 8.

Furthermore, we obtain the control set (u∗1 , u∗2 , u∗3 , u∗4 , u∗5 ) characterized by

u∗1 (t ) = max{0, min(1, 1 )},

u∗2 (t ) = max{0, min(1, 2 )},
u∗3 (t ) = max{0, min(1, 3 )},
u∗4 (t ) = max{0, min(1, 4 )},
u∗5 (t ) = max{0, min(1, 5 )},

where,
γ (Ip + ϒ It p )(Sh1 − Sh2 + It h3 − It h4 )
1 = −
Nw1
Bv (Sh1 − Sh3 + I p h2 − I p h4 )
2 = −
(k + B )w2
I p h2 − I p h5 + h4 ϒ It p − ϒ It p h5 − h7 ϒ It p
3 =
w3
(h4 − h6 − h7 )It p + It (h3 − h6 )
4 =
w4
h8 B
5 =
w5
Proof. By using Pontryagin’s maximum principle we obtain the following system of adjoint variables:
 
dh1 dH (1 − u2 )Bv γ (Ip + ϒ It p )(1 − u1 ) h2 γ (I p + ϒ It p )(1 − u1 ) h3 (1 − u2 )Bv
=− = −h1 − − −μ − −
dt dS k+B N N k+B
h1 γ (1 − u1 )S
γ 1 − u S
h3 γ (1 − u1 )It
dh2 dH ( 1) (1 − u2 )Bv
=− = −b1 + − h2 − − β1 − u3 − α1 − μ +
dt dI p N N k+B N
1 − u Bv γ (1 − u1 )It

( 2)
−h4 + − h5 (β1 + u3 )
k+B N
 
dh3 dH γ (Ip + ϒ It p )(1 − u1 )
= = −b2 − h3 − − β2 − u4 − α2 − σ1 − μ
dt dIt N
h4 γ (I p + ϒ It p )(1 − u1 )
− − h6 (β2 + u4 ) − h8 σ1
N
dh4 dH h1 γ (1 − u1 )S h2 γ (1 − u1 )S h3 γ (1 − u1 )It
=− = −b3 + − +
dt dIt p N N N
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 451

γ 1 − u I

( 1) t
−h4 − σ − u3 − u4 − α1 − α2 − σ2 − μ
N
−h5 (σ e + u3 ) − h6 (σ (1 − e ) + u4 ) − h7 (σ (1 − g)(1 − e ) + u3 + u4 ) − h8 σ2
dh5 dH dh6 dH
=− = −h1 δ1 − h5 (−δ1 − μ ) =− = −h1 δ2 − h6 (−δ2 − μ )
dt dR p dt dRt
dh7 dH
=− = −h1 δ3 − h7 (−δ3 − μ )
dt dRt p
dh8 dH h1 (1 − u2 )Bv S h2 (1 − u2 )Bv I p h3 (1 − u2 )Bv S h4 (1 − u2 )Bv I p
=− =− − + + − h8 (−u5 − μb )
dt dB (k + B )2 (k + B )2 (k + B )2 (k + B )2
∂H ∗
Similarly to get the controls ,we solved the equation, ∂ ui = 0 at ui , for i = 1, . . . , 5 and obtained:

γ (Ip + ϒ It p )(Sh1 − Sh2 + It h3 − It h4 )

u∗1 = −
Nw1
Bv (Sh1 − Sh3 + ϒ I p h2 − I p λ4 )
u∗2 =−
(k + B )w2
I p h2 − I p h5 + h4 ϒ It p − ϒ It p h5 − h7 ϒ It p
u∗3 =
w3
( h 4 − h 6 − h 7 )ϒ It p + It (h3 − h6 )
u∗4 =
w4
h8 B
u∗5 =
w5
When we write by using standard control arguments involving the bounds on the controls, we conclude:
1 if 0 < 1 < 1
u∗1 = 0 if 1 ≤ 0
1 if 1 ≥ 1.
2 if 0 < 2 < 1
u∗2 = 0 if 2 ≤ 0
1 if 2 ≥ 1.
3 if 0 < 3 < 1
u∗3 = 0 if 3 ≤ 0 .
1 if 3 ≥ 1
4 if 0 < 4 < 1
u∗4 = 0 if 4 ≤ 0 .
1 if 4 ≥ 1
5 if 0 < 5 < 1
u∗5 = 0 if 5 ≤ 0 .
1 if 5 ≥ 1
In compact notation
u∗1 (t ) = max{0, min(1, 1 )},
u∗2 (t ) = max{0, min(1, 2 )},
u∗3 (t ) = max{0, min(1, 3 )},
u∗4 (t ) = max{0, min(1, 4 )},
u∗5 (t ) = max{0, min(1, 5 )}.
γ (Ip + ϒ It p )(Sh1 − Sh2 + It h3 − It h4 )
1 = −
Nw1
Bv (Sh1 − Sh3 + I p h2 − I p h4 )
2 = −
(k + B )w2
I p h2 − I p h5 + h4 ϒ It p − ϒ It p h5 − h7 ϒ It p
3 =
w3
(h − h6 − h7 )ϒ It p + It (h3 − h6 )
4 = 4
w4
452 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

h8 B
5 =
w5
The optimality system is formed from the optimal control system (the state system) and the adjoint variable system by
incorporating the characterized control set and initial and transversal condition.
⎧ dS
⎪
⎪ = π + δ1 R p + δ2 Rt + δ3 Rt p − ((1 − u∗1 )λ p + (1 − u∗2 )λt + μ )S
⎪ dt
⎪
⎪
⎪
⎪
⎪ dI p
⎪
⎪ = (1 − u∗1 )λ p S − (1 − u∗2 )λt I p − (β1 + u∗3 )I p − (α1 + μ )I p
⎪
⎪ dt
⎪
⎪
⎪
⎪
⎪ dIt = (1 − u∗2 )λt S − (1 − u∗1 )λ p It − (β2 + u∗4 )It − (α2 + σ1 + μ )It
⎪
⎪
⎪ dt
⎪
⎪
⎪ dIt p
⎪
⎪
⎪ dt = (1 − u2 )λt Ip + (1 − u1 )λ p It − (σ + u3 + u4 )It p − (α1 + α2 + σ2 + μ )It p
∗ ∗ ∗ ∗
⎪
⎪
⎪
⎪
⎪
⎪
⎪ dR p
= (β1 + u∗3 )I p + (σ e + u∗3 )It p − (δ1 + μ )R p
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt
= (β2 + u∗4 )It + (σ g(1 − e ) + u∗4 )It p − (δ2 + μ )Rt
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dRt p
⎪
⎪ = (σ (1 − g)(1 − e ) + u∗3 + u∗4 )It p − (δ3 + μ )Rt p
⎪
⎪ dt
⎪
⎪
⎪
⎪ dB
⎪
⎪ = Q + σ1 It + σ2 It p − (u∗5 + μb )B
⎪ dt
⎪
⎪
⎪  ! " ! "  ! " ! "
⎪
⎪ 1 − u∗2 Bv γ (Ip + ϒ It p ) 1 − u∗1 h2 γ (I p + ϒ It p ) 1 − u∗1 h3 1 − u∗2 Bv
⎪
⎪ dh1
= −h1 − − −μ − −
⎪
⎪ k+B k+B
⎪
⎨
dt N N
 ! " ! "  ! "
dh2 h1 γ (1−u∗1 )S γ 1 − u1 S ∗ ∗
1 − u2 Bv h3 γ 1 − u∗1 It
⎪
⎪ = −b1 + − h2 − − β1 − u3 − α1 − μ +
∗
⎪
⎪ dt
N
N k+B N
⎪
⎪  ! " ! " 
⎪
⎪
⎪
⎪
∗
1 − u2 Bv γ 1 − u1 It∗ ! "
⎪
⎪ −h4 + − h5 β1 + u∗3
⎪
⎪ k + B N
⎪
⎪  ! "  ! "
⎪
⎪
⎪
⎪ dh3 γ (Ip + ϒ It p ) 1 − u∗1 h4 γ (I p + ϒ It p ) 1 − u∗1 ! "
⎪
⎪ = −b2 − h3 − − β2 − u4 − α2 − σ1 − μ −
∗
− h6 β2 + u∗4 − h8 σ1
⎪
⎪ dt N N
⎪
⎪ ! " ! " ! "  ! " 
⎪
⎪
⎪
⎪ dh4 h1 γ 1 − u∗1 S h2 γ 1 − u∗1 S h3 γ 1 − u∗1 It γ 1 − u∗1 It
⎪
⎪ = −b3 + − + − h4 − σ − u3 − u4 − α1 − α2 − σ2 − μ
∗ ∗
⎪
⎪ dt N N N N
⎪
⎪ ! " ! "
⎪
⎪ −h5 (σ e + u3 ) − h6 σ (1 − e ) + u∗4 − h7 σ (1 − g)(1 − e ) + u∗3 + u∗4 − h8 σ2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ dh5
= −h1 δ1 − h5 (−δ1 − μ )
⎪
⎪
⎪
⎪ dt
⎪
⎪
⎪
⎪ dh6 = −h1 δ2 − h6 (−δ2 − μ )
⎪
⎪
⎪
⎪ dt
⎪
⎪ dh7
⎪
⎪
⎪ dt = −h1 δ3 − h7 (−δ3 − μ )
⎪
⎪
⎪
⎪ ! " ! " ! " ! "
⎪
⎪ h 1 − u∗2 Bv S h2 1 − u∗2 Bv I p h3 1 − u∗2 Bv S h4 1 − u∗2 Bv I p ! "
⎪
⎩ dh8 = − 1 − + + − h8 −u∗5 − μb
dt 2 2 2 2
(k + B ) (k + B ) (k + B ) (k + B )
(27)

hi (t f ) = 0, i = 1, 2, 3, S(0 ) = S0 , V (0 ) = V0 , C (0 ) = C0 , I (0 ) = I0 , and R(0 ) = R0 . 

4.2. Uniqueness of the optimality system

Due to the boundedness of the model (27) which is considered the state , adjoint functions and also considering Lipschitz
structure of the ordinary differential equations. Then it is possible to show the uniquines of the resulting optimality system
by considering small time interval. Hence the following theorem:

Theorem 6. For t ∈ [0, tf ], the bounded solutions to the optimality system are unique.

For the proof of the theorem See [18].

 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 453

Table 2
Parameter values for Pneumonia model.

Parameter symbol Value Source

δ1 0.002–0.1 [19]
δ2 0.00904–0.99 [20]
δ3 0.01 Assumed
γ 0.001 [19]
v 0.9 Assumed
k 50.0 0 0 [21]
β1 0.002 Estimated
α1 0.051 Estimated
β2 0.002 Estimated
α2 0.0052 Estimated
σ 0.1 Assumed
g 0.5–1 Assumed
e 0.5–1 [22]
μ 0.00247 [19]
σ1 0.8 Assumed
σ2 0.9 Assumed
μb 0.0 0 0 0125 [21]

5. Numerical simulation

In this section, we performed numerical simulation of the co-infection model (1) and the resulting optimality system
(27). We make use of the parameter values given in Table 2 for the simulation.
To simulate the model, which is for obtaining the optimal solution of the resulting optimality system iterative technique
is applied. By considering the initial condition of the state system and the final condition of the adjoint systems, we use
forward fourth-order Runke–Kutta method to solve the state system and backward fourth-order Runge–Kutta method for
solving the state system. The adjoint systems is solved by using the initial guess of the controls incorporating with the
obtained solution for the state system. The controls continues to be updated by combining from the previous result of the
controls with the characterization. The solution of the state and adjoint system is repeated by the updated controls. This
condition continues repeatedly upto consecutive iteration are close enough each other, [23].
We proposed the following five strategies for numerical simulation of the co-infection model:
(i). Prevention effort for Pneumonia and treatment effort for Typhoid fever disease (u1
= 0, u4
= 0, u2 = u3 = u5 = 0).
(ii). Using prevention effort for Typhoid fever disease and treatment effort for Pneumonia disease ((u2
= 0, u3
= 0, u5
=
0, u1 = 0, u4 = 0 )).
(iii). Using treatment effort for both disease (u3
= 0, u4
= 0, u1 = u2 = u5 = 0).
(iv). Using all the intervention efforts (u1
= 0, u2
= 0, u3
= 0, u4
= 0, u5
= 0).
We assumed b1 = 25, b2 = 25, b3 = 25, w1 = 4, w2 = 3, w3 = 5, w4 = 6 and w5 = 7 for simulation of the model with op-
timal control and also for cost-effectiveness analysis. Additionally we used S(0 ) = 10 0 0, I p (0 ) = 300, It (0 ) = 200, It p (0 ) =
150, R p (0 ) = 200, Rt (0 ) = 150, Rt p (0 ) = 150 and B(0 ) = 0 as initial values.

5.1. Control with prevention of Pneumonia disease and treatment of Typhoid fever disease

The simulation results from Figs. 2 and 3 shows that a control with prevention of Pneumonia disease and treatment of
Typhoid fever disease have a potential of decreasing the co-infectious, Pneumonia infectious and Typhoid fever infectious
populations. From this we conclude that applying an optimized controls (prevention of Pneumonia disease and treatment of
Typhoid fever disease) can eradicate both diseases from the community in a specified period of time.

5.2. Control with prevention effort for Typhoid fever and treatment effort for Pneumonia.

Here we used prevention for Typhoid fever disease and treatment for Pneumonia disease as intervention strategy, and
Figs. 4 and 5 shows that, the number of co-infectious, Pneumonia infectious and Typhoid fever infectious population goes
down in the specified time. Therefore, this strategies is effective in eradicating the diseases from the community in a speci-
fied period of time.

5.3. Control with treatment effort only for both disease

In this strategy, we used treatment as an intervention for both Pneumonia and Typhoid fever disease. The simulation
results in Fig. 6 and 7 shows that the number of co-infectious, Pneumonia infectious and Typhoid fever infectious population
goes down in the specified time. Therefore, this strategies is effective in eradicating the diseases from the community in a
specified period of time.
454 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

Fig. 2. Simulation of optimal control with prevention of Pneumonia disease and treatment of Typhoid fever disease.

Fig. 3. Simulation of optimal control with prevention of Pneumonia disease and treatment of Typhoid fever disease.

Fig. 4. Simulation of optimal control with prevention of Typhoid fever disease and treatment of Pneumonia disease.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 455

Fig. 5. Simulation of optimal control with prevention of Typhoid fever disease and treatment of Pneumonia disease.

Fig. 6. Simulation of optimal control with treatment of both Pneumonia and Typhoid fever diseases.

5.4. Control with all intervention efforts

Here we used all interventions efforts (prevention of both disease and treatment of both disease). treatment and screen-
ing controls as intervention. From the simulation results in Figs. 8 and 9 we observe that optimal control of the combination
of all interventions helps to bring down the co-infectious, Pneumonia infectious as well as the Typhoid fever infectious pop-
ulations. Therefore, applying this strategy is can eradicate the diseases from the community.

6. Cost-effectiveness analysis

Cost-effectiveness analysis used to rank the implemented strategies interims of their cost. To achieve, this we used in-
cremental cost-effectiveness ratio (ICER), stated by Tilahun et al. [15];
Difference in costs between strategies
“ICER = .”
Difference in health effects between strategies
In Table 3, we obtain the total number of co-infectious averted and total cost for the implemented strategies. The total
number of co-infectious averted is obtained from the differences of total individuals after and before the implementation
of the control strategies. And also to find the total cost for the implemented strategies we used the cost function, which is
1 2 1 2 1 2 1 2 1 2
2 w1 u1 , 2 w2 u2 , 2 w3 u3 , 2 w4 u4 and 2 w5 u5 over time. We used the parameter values in Table 2 and to apply ICER technique
first we ordered the intervention strategies for pairwise comparison as in Table 3 from strategy 1 to 4 with increasing order
of effectiveness.
456 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

Fig. 7. Simulation of optimal control with treatment of both Pneumonia and Typhoid fever diseases.

Fig. 8. Simulations of optimal control with all interventions.

Table 3
Number of co-infectious averted and total cost of each strategies.

Strategies Description Averted Total cost (USD)

1 Pneumonia prevention and Typhoid fever treatment 370,405 1354.5

2 Typhoid fever prevention and Pneumonia treatment 370,792 1204
3 Control with treatment effort only for both disease 375,659 1655.5
4 Control with all interventions 375,911 2709

First we compared the cost effectiveness of strategy 1 and 2.

1, 354.5
ICER(1 ) = = 0.003,
370, 405
1, 354.5 − −1204
ICER(2 ) = = −0.389.
370, 405 − 370, 792

From ICER (1) and ICER (2) we can see that strategy 2 saves 0.389 than strategy 1. Therefore, we exclude strategy 1,
because it is a bit expensive, so we continue to compare strategy 2 and 3.
 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 457

Fig. 9. Simulation of optimal control with all interventions.

Fig. 10. Cost function of the intervention strategies.

1, 204
ICER(2 ) = = 0.00325,
370, 792
1204 − −1655.5
ICER(3 ) = = 0.093.
370, 792 − 355, 659
Similarly, from ICER (2) and ICER (3) we can see that strategy B saves 0.00325 than strategy 3. Therefore, we exclude
strategy 3, because it is a bit expensive and we continue to compare strategy 2 and 4.

ICER(2 ) = 0.00325
1204 − 2, 709
ICER(4 ) = = 0.029
370, 792 − 375, 911

Similarly, from ICER (2) and ICER (4) we can see that strategy 2 saves 0.00325 than strategy 4. Therefore, we exclude
strategy 4, because it is a bit expensive. Therefore, we conclude that strategy 2 (Typhoid fever prevention and Treatment of
Pneumonia) is the cheapest of all compared strategies, that meant it is the most cost-effective for Pneumonia–Typhoid fever
disease co-infection control interventions strategy but strategy 4 is the most expensive of all the above strategies .
For further elaboration we plotted the cost function of each strategy in Fig. 10 and the figure shows that strategy 2
(Pneumonia treatment and Typhoid fever prevention) is the least cost for implementing the intervention and strategy 4
(using all intervention) costs the highest of all strategies for implementation.
458 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459

7. Discussions and conclusions

In Section 2, we briefly described and proposed a Pneumonia–Typhoid fever co-infection model, which is deterministic
in its nature and also the population is assumed to be variable in size. In Section 3, we analyzed the model by obtaining the
feasible region, positivity of the solution set, basic reproductive number which is the maximum of the basic reproductive
number of Pneumonia and Typhoid fever only models, equilibria points and their stability are analysed. Moreover, by using
[13] theorem we investigated bifurcation of the model to explore the possibility of backward or forward bifurcation. The
impact of the two diseases on each other are analysed and we showed that Pneumonia cases increase Typhoid fever cases
and also Typhoid fever cases increase Pneumonia cases. Moreover, we concluded that treatment of Pneumonia have an
impact of reducing Typhoid fever disease. Sensitivity analysis of basic parameters and interpretation of the sensitivity index
is also done in Section 3. In Section 4, we extended the full Pneumonia–Typhoid fever co-infection model by applying
optimal control interventions and we obtained the Hamiltonian, the adjoint variables , the characterization of the controls
and the optimality system. In Section 5, we simulate the optimality system by considering different strategies as follows:

• By applying prevention for Pneumonia and treatment for Typhoid fever disease.
• By applying prevention for Typhoid fever and treatment for Pneumonia disease.
• By applying treatment effort for both disease.
• By applying all control interventions.

In Section 6, numerically we investigated cost effectiveness analysis to determine, the least and the most expensive
strategies by using ICER technique. From the pairwise comparison result we conclude that, applying prevention effort for
Typhoid fever disease and treatment effort for Pneumonia disease is the best strategies interims of cost as well as health
benefits.

Acknowledgement

We would like to express our heartfelt appreciation to Pan African University Institute of Basic Sciences Technology
and Innovation (PAUSTI) for financial support and also we are grateful to the anonymous reviewers for their constructive
comments, which have improved the manuscript.

Appendix A

Theorem 4.1 of [13] Consider the following general system of ODEs with a parameter φ .

dx
= f (x, φ ), f : Rn × R → R and f ∈ C2 (Rn × R ), (28)
dt

point of
the system (that is, f(0, φ ) ≡ 0, ∀φ ) and assume
where 0 is an equilibrium
∂f
A1: A = Dx f (0, 0 ) = ∂ x i (0, 0 ) is the linearization matrix of (28) around the equilibrium point 0 with φ evaluated at 0.
j
Zero is a simple eigenvalue of A and other eigenvalues of A have negative real parts;
A2: Matrix A has a right eigenvector w and a left vector v (each corresponding to the zero eigenvalue).
Let fk be the kth component of f and


n
∂ 2 fk
a= vk wi w j ( 0, 0 ), (29)
k, i, j=1
∂ xi ∂ x j


n
∂ 2 fk
b= vk wi ( 0, 0 ). (30)
k, i=1
∂ xi ∂ φ

The local dynamics of system (28) around 0 is totally determined by the signs of a and b:

i. a > 0, b > 0. When φ < 0 with |φ | ≤ 1, 0 is locally asymptotically stable, and there exists a positive unstable equilibrium;
when 0 < φ 1, 0 is unstable and there exists a negative and locally asymptotically stable equilibrium;
ii. a < 0, b < 0. When φ < 0 with |φ | 1, 0 is unstable; when 0 < φ 1, 0 is locally asymptotically stable, and there exists a
positive unstable equilibrium;
iii. a > 0, b < 0. When φ < 0 with |φ | 1, 0 is unstable, and there exists a locally asymptotically stable negative equilibrium;
when 0 < φ 1, 0 is stable, and a positive unstable equilibrium appears;
iv. a < 0, b > 0. When φ changes from negative to positive, 0 changes its stability from stable to unstable. Correspondingly a
negative unstable equilibrium becomes positive and locally asymptotically stable.

Particularly, if a > 0 and b > 0, then a backward bifurcation occurs at φ = 0.

 G.T. Tilahun et al. / Applied Mathematics and Computation 316 (2018) 438–459 459

References

[1] McLuckie, Leach, Respiratory Disease and its Management, Springer, 2009, p. 51.
[2] B. Greenwood, The Epidemiology of Pneumococcal Infection in Children in the Developing World, The Royal Society, 20 0 0.
[3] WHO, Technical Basis for the WHO Recommendations on the Management of Pneumonia in Children at First Level Health Facilities, WHO, 2008.
[4] G.T. Tilahun, O.D. Makinde, D. Malonza, Modelling and optimal control of pneumonia disease with cost-effective strategies, J. Biol. Dyn. 11 (Sup2)
(2017) 400–426. PMID:28613986.
[5] M. Kgosimore, G.R. Kelatlhegile, Mathematical analysis of typhoid infection with treatment, J. Math. Sci. Adv. Appl. 40 (2016) 75–91.
[6] O. David Onyinge, N. Omolo Ongati, Francis Odundo mathematical model for coinfection of pneumonia and HIV/AIDS with treatment, Int. J. Sci. Eng.
Appl. Sci. (IJSEAS) 2 (1) (2016).
[7] S. Edward, N. Nyerere, Modelling typhoid fever with education, vaccination and treatment, Eng. Math. 1 (1) (2016) 44–52, doi:10.11648/j.engmath.
20160101.14.
[8] O.C. Akinyi, J.Y. Mugisha, A. Manyonge, C. Ouma, Modelling the impact of misdiagnosis and treatment on the dynamics of malaria concurrent and
co-infection with pneumonia, Appl. Math. Sci. 7 (2013) 6275–6296.
[9] S. Mushayabasa, C.P. Bhunu, N.A. Mhlanga, Modeling the transmission dynamics of typhoid in malaria endemic settings, Appl. Appl. Math. Int. J. (AAM)
9 (1) (2014) 121–140. ISSN:1932-9466. http://pvamu.edu/aam.
[10] K.R. Adeboye, M. Haruna, A mathematical model for the transmissionand control of malaria and typhoid co-infection using sirs approach, Niger. Res.
J. Math. 2 (2015) 1–24.
[11] O.C. Akinyi, J.Y. Mugisha, A. Manyonge, C. Ouma, A model on the impact of treating typhoid with anti-malarial: dynamics of malaria concurrent and
co-infection with typhoid, Appl. Math. Sci. 9 (2015) 541–551.
[12] P. van den Driessche, J. Watmough, Reproduction numbers and subhreshold endemic equilibria for compartmental models of disease transmission,
Math. Biosci. 180 (2002) 29–48.
[13] C. Castillo-Chavez, B. Song, Dynamical models of tuberculosis and their applications, Math. Biosci. Eng. 1 (2) (2004) 361–404.
[14] N. Chitnis, J.M. Cushing, J.M. Hyman, Bifurcation analysis of a mathematical model for malaria transmission, SIAM J. Appl. Math. 67 (2006) 24–45.
[15] G.T. Tilahun, O.D. Makinde, D. Malonza, Modelling and optimal control of pneumonia diseasewith cost-effective strategies, J. Biol. Dyn. 11 (Sup2) (2017)
1–27.
[16] L.S. Pontryagin, V.G. Boltyanskii, R.V. Gamkrelize, E.F. Mishchenko, The Mathematical Theory of Optimal Processes, John Wiley & Sons, New York,
NY,USA, 1986.
[17] W.H. Fleming, R.W. Rishel, Deterministic and Stochastic Optimal Control., Springer Verlag, New York, 1982.
[18] K.R. Fister, S. Lenhart, J.S.M. Nally, Optimizing chemotherapy in an HIV model, Electron. J. Diff. Equ. 32 (1998) 1–12.
[19] E. Joseph, Mathematical analysis of prevention and control strategies of pneumonia in adults and children, Math. Theor. Model. 5 (10) (2015) 21–39.
[20] I.A. Adetunde, Mathematical models for the dynamics of typhoid fever in Kassena–Nankana district of upper east region of Ghana, Mod. Math. Stat. 2
(2008) 45–49.
[21] M. Gosh, P. Chandra, P. Sinha, J.B. Shukla, Modelling the spread of bacterial infectious disease with environmental effect in a logistically growing
human population, Non-linear Anal. Real World Appl. 7 (2006) 341–363.
[22] D. Pessoa, Modelling the daynamics of streptococcus Pneumoniae transmmission in children, University of De Lisboa, 2010 (Master’s thesis).
[23] S. Lenhart, J.T. Workman, Optimal Control Applied to Biological Models, Chapman & Hall/CRC, London, UK, 2007. (Math. Comput. Biol. Ser.)