Sunteți pe pagina 1din 14

Mathematical Biosciences 263 (2015) 3750

Contents lists available at ScienceDirect

Mathematical Biosciences
journal homepage: www.elsevier.com/locate/mbs

Global stability of an infection-age structured HIV-1 model linking


within-host and between-host dynamics
Mingwang Shen a , Yanni Xiao a,, Libin Rong b
a
b

School of Mathematics and Statistics, Xian Jiaotong University, Xian 710049, PR China
Department of Mathematics and Statistics, Oakland University, Rochester, MI 48309, USA

a r t i c l e

i n f o

Article history:
Received 6 May 2014
Revised 30 January 2015
Accepted 5 February 2015
Available online 14 February 2015
Keywords:
HIV-1 infection
Infection-age
Nested model
Saturated function
Global asymptotical stability
Lyapunov functional

a b s t r a c t
Although much evidence shows the inseparable interaction between the within-host progression of HIV-1
infection and the transmission of the disease at the population level, few models coupling the within-host
and between-host dynamics have been developed. In this paper, we adopt the nested approach, viewing the
transmission rate at each stage (primary, chronic, and AIDS stage) of HIV-1 infection as a saturated function
of the viral load, to formulate an infection-age structured epidemic model. We explicitly link the individual and
the host population scale, and derive the basic reproduction number R0 for the coupled system. To analyze
the model and perform a detailed global dynamics analysis, two Lyapunov functionals are constructed to
prove the global asymptotical stability of the disease-free and endemic equilibria. Theoretical results indicate
that R0 provides a threshold value determining whether or not the disease dies out. Numerical simulations
are presented to quantitatively investigate the inuence of the within-host viral dynamics on between-host
transmission dynamics. The results suggest that increasing the effectiveness of inhibitors can decrease the
basic reproduction number, but can also increase the overall infected population because of a lower diseaseinduced mortality rate and a longer lifespan of HIV infected individuals.
2015 Elsevier Inc. All rights reserved.

1. Introduction
The last few decades have witnessed a substantial increase
in the application of mathematical models to HIV-1 infection at
both the population level and within-host level. On one hand, the
overwhelming majority of articles [17] were exclusively focused
on the transmission dynamics of HIV at the population level to predict the future prevalence and suggest the possible control strategies.
On the other hand, a relatively large number of publications [812]
only considered the viral dynamic models that characterize the interaction between CD4+ T cells and virus with the aim of helping to guide
treatment strategies. These two distinct but interrelated levels have
remained non-intersecting from the viewpoint of mathematical modeling. However, more empirical evidence [1318] indicated that there
was a strong relation between viral load and the disease transmission
rate, i.e., a higher circulating viral load was positively associated with
a higher rate of host-to-host transmission and a larger scale of HIV epidemics [19]. Coupling within-host viral dynamics with between-host
transmission dynamics, and further examining the impact of viral dynamics on between-host transmission dynamics remain unclear and
fall within the scope of this study.

Corresponding author. Tel.: +86 29 82663938; fax: +86 29 82663938.


E-mail address: yxiao@mail.xjtu.edu.cn, yannixiao317@hotmail.com (Y. Xiao).

http://dx.doi.org/10.1016/j.mbs.2015.02.003
0025-5564/ 2015 Elsevier Inc. All rights reserved.

Some researchers integrated the within-host model into the epidemiological model by introducing the viral load-dependent transmission rate or disease-induced mortality rate [2023]. Such a nested
approach bridging these two dynamical processes has been rst considered in [24]. Gilchrist and Coombs [20] and Coombs et al. [21] formulated a nested model to evaluate the direction of natural selection
and competition of parasites at both the within- and between-host
levels. In a comprehensive review [25], the authors outlined a general nested model assuming that the infection-age-dependent transmission rate and mortality rate are associated with the within-host
variables.
Although most of the above mentioned studies have built nested
models, they mainly considered the evolution of the host and parasite from the viewpoint of evolutionary dynamics. They did not study
the long-term dynamics of the coupled system, which is of epidemiological interest especially for public-health organizations. Our main
purpose of this study is to develop a nested model for exploring the
potential effect of within-host dynamics on the between-host transmission dynamics and establish global dynamical properties for the
nested model. Recently, Feng et al. [22] constructed a model incorporating the explicit interdependence of the viral dynamics and the
between-host dynamics and found bistability of their new coupling
system. Their viral dynamic model was formulated without describing the variation in the transmission rate or disease-induced mortality

38

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

rate at different disease progression stages. Martcheva and Li [26]


built a nested two-strain model of HIV infection in the case of superinfection and discussed the effect of within-host infection on the
population level of disease. Numfor et al. [27] formulated a coupled
within-host and between-host model to minimize the number of free
virions, infected individuals and the cost of implementing the control.
We propose an age-of-infection model in the form of partial differential equations (PDEs) to represent the staged progression nature of HIV disease at the population level. Viral load-dependent
transmission rates are used in different disease progression stages
to link the within-host viral dynamic model and the transmission
dynamic model at the population level. Specically, we adopt a saturation function to describe the relationship between viral load and
the transmission rate in each disease progression stage. The global
dynamical behavior of our proposed system is studied by developing
suitable Lyapunov functionals [11,2830]. Finally we investigate to
which parameter the basic reproduction number is most sensitive and
how the within-host parameters impact the total number of infected
individuals.
2. Linking the within- and between-host models
2.1. Within-host model
The classic HIV dynamic model, widely used to describe the plasma
viral load changes in HIV infected individuals, has three state variables: T, the concentration of uninfected target T cells; T , the concentration of productive infected T cells; and V, the concentration
of free virus particles in the blood. The model equations [8] are as
follows

dT (a)

= uT (a) kT (a)V (a)

da


dT (a)
= kT (a)V (a) qT (a)

da

dV (a)

= pT (a) c1 V (a)
da

T = (R0V

c1 u
,
1)
kp

V = (R0V

T (a2 ) = T,

T (a2 ) = T ,

V2 (a2 ) = V1 ,

T and V1 are given by (2). If we let V1 (a) be the concentration


where T,
of free virus particles in the blood before the stage of AIDS (a a2 ),
then we can write


V (a) =

V1 (a),

0 a a2 ,

V2 (a),

a2 a a3 .

2.2. The nested model


We consider a high risk group of sexually-active population which
is decomposed into four categories: susceptibles (S), HIV-infected individuals at the primary stage (I1 ), HIV-infected individuals at the
chronic or asymptomatic stage (I2 ), and those with fully developed
AIDS symptoms (A). Uninfected individuals are recruited into the
susceptible population at a positive constant rate b. People exit the
sexually-active population at a positive constant rate m due to natural death or behavior changes. Let t denote time and a denote the
infection age. We assume that all of the infected individuals with the
same infection age are homogeneous and have the same rates.
Let i1 (a, t), i2 (a, t) and e(a, t) be the densities of infected individuals
in I1 , I2 and AIDS classes, respectively, at time t and infection age a,
then

I1 (t) =

a1
0

i1 (a, t)da, I2 (t) =

a2

a1

i2 (a, t)da, A(t) =

u
1) .
k

(2)

Note that within-host HIV infection typically results in a vast replication of the virus during the acute or primary phase of the infection,
which is followed by a chronic phase in which the viral load approaches a lower quasi-steady state and later a sharp and sudden rise
of viral load when the immune system collapses and full symptoms
of AIDS are developed [14]. However, the above model (1) cannot describe the dynamic of HIV in the AIDS stage when a sharp and sudden
rise of viral load appears. Thus, we describe the viral dynamics within
AIDS patients by replacing the constant rate of viral clearance c1 with
a decreasing function of infection age c2 (a) in model (1), and denote
the viral load of this stage as V2 (a). This is reasonable as HIV is attacked by the immune cells and in the AIDS stage the immune system
cannot control the virus. The other parameters remain the same as
those in (1).
If we assume a1 to be the time when the chronic stage starts, a2 to
be the time when the AIDS stage starts, and a3 to be the time when
the infected individual dies because of AIDS, then the initial values of
the uninfected T cells, infected T cells and viral load at the beginning

a3
a2

e(a, t)da,
(3)

(1)

where a represents the infection age (the time since infection). The
generation rate and per capita death rate of healthy T cells are and
u, respectively. k denotes the infectivity rate, q denotes the death rate
of infected cells, p is the rate of production of virions by one infected
cells, and c1 is the clearance rate of free virus. For the system (1), there
kp
which determines whether HIV can persist
is a threshold R0V = quc
1
or not [31]. If R0V > 1, uninfected T cells, infected T cells and viral load
respectively, where
T , and V,
will stabilize at constant levels T,

qc1
,
T =
kp

of the AIDS stage are

are the numbers of infected individuals at primary, asymptomatic


and AIDS stage at time t 0. We denote the infection-age-dependent
disease-induced mortality rates in the I1 , I2 and A classes as
1 (a), 2 (a), and 3 (a), respectively. The infected individuals in
the I1 class progress to I2 class with a rate 1 (a) and then proceed
with a rate 2 (a) to the AIDS class. It is naturally assumed that
i (a), j (a), i = 1, 2, 3; j = 1, 2 are nonnegative, bounded and integrable. The probability that an infected individual in the I1 , I2 and A
class still stays in his/her original class at infection age a is given by

1 (a) = e
2 (a) = e

3 (a) = e

a
0

a

(m+1 (s)+1 (s))ds ,

a [0, a1 ],

(m+2 (s)+2 (s))ds

a [a1 , a2 ],

a1

a
a2

(m+3 (s))ds

a [a2 , a3 ].

(4)

As stated in Section 1, we adopt the saturated viral load-dependent


transmission rates 1 (a) in I1 and I2 classes and 2 (a) in the A class,
that is,

1 (V1 (a)) =

0 V1 (a)
,
1 + 0 V1 (a)

a [0, a2 ],

(5)

0 V2 (a)
,
1 + 0 V2 (a)

a [a2 , a3 ],

(6)

and

2 (V2 (a)) =

where V1 (a) and V2 (a) are the viral loads determined in model (1),
and  is the modication factor for reducing the transmission rate of
AIDS patients who have reduced sexual activities. 0 > 0 and 0 0
are saturated coecients.

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

We develop the complete nested dynamical model as follows:

a1
dS(t)

= b mS(t) S(t)
1 (V1 (a))i1 (a, t)da

dt

 a2
 a3

+
1 (V1 (a))i2 (a, t)da +
2 (V2 (a))e(a, t)da ,

a1
a2

i1 (a, t) i1 (a, t)

+
= (m+ 1 (a)+ 1 (a))i1 (a, t), 0 < a a1 ,

t
a

i2 (a, t) i2 (a, t)

+
= (m+ 2 (a)+ 2 (a))i2 (a, t), a1 < a a2 ,

t
a

e(a, t) e(a, t)

+
= (m + 3 (a))e(a, t),
a2 < a a3 ,
t
a

 a1
 a2

(
0,
t
)
=
S
(
t
)
1 (V1 (a))i1 (a, t)da +
1 (V1 (a))i2 (a, t)da
i

0
a1

 a3

(
V
(
a
))
e
(
a,
t
)
da
,
+
2
2

a2

 a1

(
a
,
t
)
=
1 (a)i1 (a, t)da,
i
2
1

 a2

2 (a)i2 (a, t)da,


e(a2 , t) =

a1

S(0) = S0 0, i1 (a, 0) = i10 (a), i2 (a, 0) = i20 (a), e(a, 0) = e0 (a),


(7)
L1+ (0, a1 ), i20 (a)

where i10 (a)


Here
L1+ is the space of functions that are nonnegative and Lebesgue integrable over the specied interval. For convenience of denotation,
we denote

B1 (t) = i1 (0, t),

L1+ (a1 , a2 ), e0 (a)

B2 (t) = i2 (a1 , t),

L1+ (a2 , a3 ).

B3 (t) = e(a2 , t).

(8)

R+ L1+ (0, a1 ) L1+ (a1 , a2 ) L1+ (a2 , a3 ) be the phase space

Let X =
for the system (7). By integrating the equations along the characteristic lines and incorporating the boundary conditions (see [32,33]),
explicit formulas for the solution of the age-structured model (7) can
be obtained as follows:

B1 (t a)1 (a),
i1 (a, t) =
1 (a)

i10 (a t)
,
1 (a t)

B2 (a1 + t a)2 (a),


i2 (a, t) =
2 (a)

i20 (a t)
,
2 (a t)

t > a,

a [0, a1 ],

t a,

a [0, a1 ],

t > a a1 ,
t a a1 ,

I2 (t) =

and

A(t) =

t > a a2 ,

a [a2 , a3 ],

t a a2 ,

a [a2 , a3 ],

a2



S(t) = emt S0 +


(14)

t > a3 a2 .

t
0

(b B1 (a))ema da ,


1 (a + t)
da,
1 (a)
 t
 a2 t
2 (a + t)
B2 (a)2 (a1 + t a)da +
i20 (a)
da,
I2 (t) =
2 (a)
0
a1
 t
 a3 t
3 (a + t)
B3 (a)3 (a2 + t a)da +
e0 (a)
da,
A(t) =
3 (a)
0
a2

I1 (t) =

B1 (a)1 (t a)da +

and

B1 (t) = S(t)

0
t

a2 t

a1
 t
0

(11)

+
B2 (t) =


0


0

a3 t

a1 t
0

i10 (a)

(15)

1 (V1 (t a))B1 (a)1 (t a)da


1 (V1 (a + t))i10 (a)

1 (a + t)
da
1 (a)

1 (V1 (a + t))i20 (a)

2 (a + t)
da
2 (a)

2 (V2 (a + t))e0 (a)

3 (a + t)
da ,
3 (a)

1 (t a)B1 (a)1 (t a)da

+
B3 (t) =

2 (V2 (a2 + t a))B3 (a)3 (a2 + t a)da

a2
t

a1 t

1 (V1 (a1 + t a))B2 (a)2 (a1 + t a)da

0
a1 t

(12)

0 t a3 a2 ,

Proof. We rst prove the existence of solution to system (7) and (3).
Let (S(t), i1 (a, t), i2 (a, t), e(a, t)) be a solution of system (7) and (3) up
to time T > 0. Then S(t), I1 (t), I2 (t), A(t) and B1 (t), B2 (t), B3 (t) satisfy
the following integral Eqs. (15) and (16) on t [0, a1 ].

t > a1 ,

t > a2 a1 ,

Proposition 2.1. There exists a unique nonnegative and bounded solution for system (7) and (3) with nonnegative initial conditions.

(10)

0 t a1 ,

(13)

e(a, t)da

a2

where 1 (a), 2 (a) and 3 (a) are dened in (4).


Hence,

a3

 t+a2

B3 (a2 + t a)3 (a)da

a2

a3
3 (a)
=
e0 (a t)
da,
+

3 (a t )
t+a

2


a3

B3 (a2 + t a)3 (a)da,

(9)

a [a1 , a2 ],

0 t a2 a1 ,

a1

a [a1 , a2 ],

 t

B1 (t a)1 (a)da

 a1
 a1

1 (a)
I1 (t) =
i1 (a, t)da =
i10 (at)
da,
+

(a t)
0
1

 a1

B1 (t a)1 (a)da,

i2 (a, t)da

a1

 t+a1

B2 (a1 + t a)2 (a)da

a1

 a2

2 (a)
=
i20 (a t)
da,
+

2 (a t )
t+a

 a2

B2 (a1 + t a)2 (a)da,

and

B3 (a2 + t a)3 (a),


e(a, t) =
3 (a)

e0 (a t)
,
3 (a t)

a2

39

1 (a + t)i10 (a)

1 (a + t)
da,
1 (a)

2 (a1 + t a)B2 (a)2 (a1 + t a)da

a2 t
a1

2 (a + t)i20 (a)

2 (a + t)
da.
2 (a)

(16)

Here, since the inequalities 0 < a1 < a3 a2 < a2 a1 hold in this


paper (see Table 1) and the conclusions for a1 t a3 a2 ,

40

M. Shen et al / Mathematical Biosciences 263 (2015) 3750


Table 1
Parameters and initial data chosen for the simulation.
Variable and parameter

Description

Initial or default values

Source

Variables
T (a)
T (a)
V1 (a)
V2 (a)
S(t)
i1 (a, t)
i2 (a, t)
e(a, t)
I1 (t)
I2 (t)
A(t)

Uninfected CD4+ cell population size at infection age a [0, a3 ]


Infected CD4+ cell population size at infection age a [0, a3 ]
Viral load at infection age a [0, a2 ]
Viral load at infection age a [a2 , a3 ]
Total number of susceptible population at time t
Age density of individuals in I1 class at time t and infection age a [0, a1 ]
Age density of individuals in I2 class at time t and infection age a [a1 , a2 ]
Age density of individuals in A class at time t and infection age a [a2 , a3 ]
Total number of patients in I1 class at time t
Total number of patients in I2 class at time t
Total number of patients in A class at time t

1200 cells /l
0
106 copies/ml
V1
9308
a+1
0.1268(a + 1)e 28
a90
0.0006(a 90)e 1095
a3136
0.0028(a 3136)e 149
83
551
58

[10]
[10]
[10]
See text
Assumed
Assumed
Assumed
Assumed
Assumed
Assumed
Assumed

Generation rate of uninfected CD4+ T cells


Death rate of uninfected CD4+ T cells
Infection rate of CD4+ T cells by virus
Death rate of infected CD4+ T cells
The rate of production of virions by infected cells
Clearance rate of free virus before AIDS stage
Clearance rate of free virus at AIDS stage (a [a2 , a3 ])
The duration of the primary or acute stage
The time from infection to AIDS
The time from infection to death
Susceptible population admission rate
Removal rate due to natural death m1 and changes in sexual behavior m2
The disease-induced death rate in I1 class
The disease-induced death rate in I2 class
The disease-induced death rate in A class (assumed a constant 3 here)
Rate of progression to asymptomatic stage (assumed a constant 1 here)
Rate of progression to AIDS stage (assumed a constant 2 here)
Modication factor in transmission rate of AIDS patients
The transmission rate in I1 and I2 class
The transmission rate in A class

30 l day
1
0.01 day
1
2.4 106 l day
1
0.35 day
1
3500 day
1
3 day
3
3 a3 a
(a a2 )
2
90 days
3139 days
3832.5 days
1
1.2 day
1
0.0462
day
365
0

1
1
day
693.5
1
1
day
90
1
1
day
3049
0.2495

Parameters

u
k
q
p
c1
c2 (a)
a1
a2
a3
b
m = m1 + m2
1 (a)
2 (a)
3 (a)
1 (a)
2 (a)


1 (a)
2 (a)

a3 a2 t a2 a1 , and t a2 a1 are the same as those for t


[0, a1 ], we only discuss the case 0 t a1 .
Conversely, if S(t), I1 (t), I2 (t), A(t) and B1 (t), B2 (t), B3 (t) are nonnegative functions that satisfy (15) and (16) on t [0, a1 ],
and (i1 (a, t), i2 (a, t), e(a, t)) are dened on L1+ (0, a1 ) L1+ (a1 , a2 )
L1+ (a2 , a3 ) by (9)(11), then it is easy to verify that (3) and every
equation of (7) are satised. Thus, according to Theorem 1 in [32],
(S(t), i1 (a, t), i2 (a, t), e(a, t)) is a solution of (7) and (3).
Applying standard xed-point arguments to (15) and (16), the
solution of (7) is unique (see Theorem 2 in [32]). It is easy to see that
i1 (a, t), i2 (a, t) and e(a, t) given by (9)(11) with nonnegative initial
conditions remain nonnegative. Further, if there exists t1 > 0 such
dS(t )
that S(t1 ) = 0 and S(t) > 0 for 0 t < t1 , then we obtain dt1 = b > 0
from the rst equation of (7). Therefore, S(t) 0 for all t 0. Note
that the total population size N(t) satises

N(t) = S(t) + I1 (t) + I2 (t) + A(t) = S(t) +


 a3
 a2
i2 (a, t)da +
e(a, t)da,
+
a1

a1

i1 (a, t)da

a2

and



dS(t)
d a1
d a2
dN(t)
=
+
i1 (a, t)da +
i2 (a, t)da
dt
dt
dt 0
dt a1

d a3
e(a, t)da
+
dt a2
 a1
= bmN(t)i1 (a1 , t)i2 (a2 , t)e(a3 , t)
1 (a)i1 (a, t)da
0
 a3
 a2
2 (a)i2 (a, t)da
3 (a)e(a, t)da

a1

b mN(t).

a2

[37]
[10]
[10]
[10]
[38]
[10]
See text
[5]
[7]
[36]
Assumed
[2,3]
[6]
See text
See text
See text
See text
[39]
See text
See text

We can easily get lim supt+ N(t) b/m. So the nonnegative solution for the system (7) with nonnegative initial conditions are nonnegative and bounded, which indicates that our model is well-posed.
This completes the proof.
We derive the basic reproduction number R0 for the spread of
disease at the population level. The average number of secondary
infections produced by an infected individual in the I1 class can be
b
A1 , where
expressed as RI1 = m

A1 =

a1
0

1 (V1 (a))1 (a)da.

(17)

The average number of secondary infections produced by an infected


b
A2 A3 , where
individual in the I2 class can be expressed as RI2 = m

A2 =

a1
0

1 (a)1 (a)da

(18)

is the probability that a newly infected individual survives the primary


stage and proceeds to the asymptomatic stage, and

A3 =

a2
a1

1 (V1 (a))2 (a)da.

(19)

b
Here the product m
A3 represents the expected number of new infections caused by a single newly infected individual at the asymptomatic stage. Similarly, the average number of secondary infections
produced by an infected individual in the A class can be expressed as
b
A2 A4 A5 , where
RA = m

A4 =

a2
a1

2 (a)2 (a)da, A5 =

a3

a2

2 (V2 (a))3 (a)da.

(20)

Obviously, Ai , i = 1, . . . , 5 are positive and nite. Therefore, if a typical


infected individual is introduced into a completely susceptible population, then the average number of secondary infections produced

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

during the three stages can be expressed by

R0 = RI1 + RI2 + RA =

where

b
(A1 + A2 A3 + A2 A4 A5 ).
m

(21)

For convenience of denotation, we write the transmission rates as


1 (a) and 2 (a) in the remaining theoretical analysis. Next we nd
the steady state of model (7). Model (7) has two possible equilibria.
0, 0, 0), where S = b , always exThe disease-free equilibrium E = (S,
m
ists. The endemic steady state E = (S , i1 (a), i2 (a), e (a)) satises the
following equations

 a1
 a2

b mS S

1 (a)i2 (a)da
1 (a)i1 (a)da +

0
a1

 a3

+
2 (a)e (a)da = 0,

a2

d i (a) = (m + 1 (a) + 1 (a))i (a), 0 < a a1 ,

da 1

i (a) = (m + 2 (a) + 2 (a))i2 (a), a1 < a a2 ,

da 2

d
e (a) = (m + 3 (a))e (a),
a2 < a a3 ,

da

 a1
 a2

(
0
)
=
S

(
a
)
i
(
a
)
da
+
1 (a)i2 (a)da
i

1
1
1

0
a1

 a3

(
a
)
e
(
a
)
da
,

a2

 a1

i
(
a
)
=
1 (a)i1 (a)da,
1

2
0

 a2

e (a2 ) =
2 (a)i2 (a)da.

i2 (a) =

bA2
(R0 1)2 (a),
R0

and

L3 (t) =

a2
a1

i2 (a)i2 (a, t)da, L4 (t) =

a1
0

i1 (a)i1 (a, t)da,

a3
a2

(m+2 (l)+2 (l))dl ds,

s
S 2 (s)e a (m+3 (l))dl ds.

Obviously, we have

Calculating the derivative of L1 (t) along (7), we obtain

S
1
S(t)

(mS mS(t) i1 (0, t))

a1

a2

L2 (t) =

t
0


=

t
0

(23)

i1 (a)B1 (ta)1 (a)da+

a1 t
0

i1 (t + r)i10 (r)

Thus, we have

dL2 (t)
=
dt

a1
t

i1 (a)i10 (at)

1 (a)
da
1 (at)

i1 (t r)B1 (r)1 (t r)dr

1 (t + r)
dr.
1 (r)


i1 (0)B1 (t)+ i1 (tr)B1 (r)1 (tr)dr i1 (tr)B1 (r)


0



m + 1 (t r) + 1 (t r) 1 (t r)dr
 a1 t
(t + r)
i1 (t + r)i10 (r) 1
dr
+
1 (r)
0
 a1 t


i1 (t + r)i10 (r) m + 1 (t + r) + 1 (t + r)

1 (t + r)
dr

1 (r)
= R0 B1 (t) +

i1 (a)B1 (t a)1 (a)da


0

i1 (a)

B1 (t a)(m + 1 (a) + 1 (a))1 (a)da


 a1
 a1
(a)
i1 (a)i10 (a t) 1
da
i1 (a)i10 (a t)
+
1 (a t)
t
t
(m + 1 (a) + 1 (a))
= R0 B1 (t) +

e (a)e(a, t)da.

When 0 t a1 , using (9), L2 (t) becomes

Proof. We construct a Lyapunov functional L(t) = L1 (t) + L2 (t) +


L3 (t) + L4 (t) which is continuously differentiable and nonnegative.
Here we dene

a3

s

 a1
m(S(t) S )2
i1 (0, t) + S
1 (a)i1 (a, t)da
S(t)
0

 a3
 a2
1 (a)i2 (a, t)da +
2 (a)e(a, t)da .
+

Theorem 3.1. The disease-free equilibrium E is globally asymptotically


stable if R0 1 and unstable if R0 > 1.

L2 (t) =

e (a) =

(S 1 (s) + e (a2 )2 (s))e

(m+1 (l)+1 (l))dl ds,

In this section, we investigate the stability of the disease-free equilibrium, which will be shown to be dependent on the basic reproduction number R0 .

S(t)
,
S

dL1 (t)
=
dt

3. Global stability of the disease-free equilibrium

L1 (t) = S(t) S S ln

a2

(24)
(22)

b
(R0 1)1 (a),
R0

bA2 A4
(R0 1)3 (a).
R0

i2 (a) =

s

i1 (a) = S 1 (a) i2 (a1 )1 (a) + (m + 1 (a) + 1 (a))i1 (a),




i2 (a) = S 1 (a) e (a2 )2 (a) + m + 2 (a) + 2 (a) i2 (a),
e (a) = S 2 (a) + (m + 3 (a))e (a).

i1 (a) = (b mS )1 (a)

e (a) =

(S 1 (s) + i2 (a1 )1 (s))e

where Ai , i = 1, . . . , 5 are given by (17)(20). We calculate the derivatives of i1 (a), i2 (a) and e (a) as follows:

Tedious calculation from (22) shows that if R0 > 1, the unique endemic steady state E = (S , i1 (a), i2 (a), e (a)) exists, and it is given
by

a1

i1 (0) = S (A1 + A2 A3 + A2 A4 A5 ) = R0 ,
5,
i2 (a1 ) = S (A3 + A4 A5 ), e (a2 ) = SA

a1

1
S
S =
=
,
A1 + A2 A3 + A2 A4 A5
R0

i1 (a) =

41


0

a1

1 (a)
da
1 (a t)

(i1 (a) m + 1 (a) + 1 (a)

i1 (a))i1 (a, t)da.

42

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

a
t
When t > a1 , L2 (t) = 0 1 i1 (a)B1 (t a)1 (a)da = ta i1 (t r)B1
1
(r)1 (t r)dr. Differentiating L2 (t), we have

dL2 (t)
=
dt

i1 (0)B1 (t) +


t
ta1

i1 (t r)B1 (r)1 (t r)dr


ta1

i1 (tr)B1 (r) m+ 1 (tr) + 1 (tr) 1 (tr)dr

 a1
 a1
= R0 B1 (t)+
i1 (a)B1 (ta)1 (a)da
i1 (a)B1 (t a)
0
0


m + 1 (a) + 1 (a) 1 (a)da
 a1
= R0 B1 (t)+
(i1 (a) (m+ 1 (a)+ 1 (a))i1 (a))i1 (a, t)da.
0

which is the same as the expression obtained when 0 t a1 . Hence,


for all t 0, the above equality holds.
Similarly, using (10) and (11), we obtain

dL3 (t)
=
dt

i2 (a1 )B2 (t)




a2

a1

(i2 (a) (m + 2 (a) + 2 (a))i2 (a))i2 (a, t)da,

and

dL4 (t)
= e (a2 )B3 (t) +
dt
Combining equalities of
and (8), we get

a3
a2

e (a) (m + 3 (a))e (a) e(a, t)da.

dL1 (t) dL2 (t) dL3 (t)


, dt , dt
dt

and

dL4 (t)
,
dt

and using (7)

4. Global stability of the endemic equilibrium


In Theorem 3.1 we have shown that if R0 > 1, the infection-free
equilibrium is unstable. The disease may spread even when a single
infection is introduced into the population. The permanence of system
(7) can be obtained by using similar methods to those in [29,34,35].
Now we assume R0 > 1, and show that the unique endemic equilibrium E is globally asymptotically stable by constructing an appropriate Lyapunov functional. To complete our proof, we need the
following lemma.
Lemma 4.1. Each solution of system (7) satises the following equalities:

S(t) i1 (a, t) i1 (0)


da
S i1 (a) i1 (0, t)
0

 a2
S(t) i (a, t) i1 (0)
1 (a)i2 (a) 1 2
+S
da
S i2 (a) i1 (0, t)
a1

 a3
S(t) e(a, t) i1 (0)
2 (a)e (a) 1
+ S
da = 0.
S e (a) i1 (0, t)
a2

 a1
i (a, t) i2 (a1 )
(ii)
1 (a)i1 (a) 1 1
da = 0.
i1 (a) i2 (a1 , t)
0

 a2
i (a, t) e (a2 )
(iii)
2 (a)i2 (a) 1 2
da = 0.
i2 (a) e(a2 , t)
a1

(i) S

= S

a1

a2

as

1 (a)i1 (a)da+

a2
a1

1 (a)i2 (a)da+

a2
a1

a3
a2

2 (a)e (a)da

1 (a)i1 (a) 1

Similarly, we have

dL(t)
dt

a1

S(t) i1 (a, t) i1 (0)


da
S i1 (a) i1 (0, t)
0

 a2
S(t) i (a, t) i1 (0)
1 (a)i2 (a) 1 2
+ S
da
S i2 (a) i1 (0, t)
a1

 a3
S(t) e(a, t) i1 (0)
2 (a)e (a) 1
+ S
da.
S e (a) i1 (0, t)
a2

= S

+e (a2 )2 (a))i2 (a, t)da


 a3
+
(S 2 (a) + e (a) (m + 3 (a))e (a))e(a, t)da

Therefore, R0 1 implies that

 a1
 a2
i (0)
1
S(t)
1 (a)i1 (a, t)da +
1 (a)i2 (a, t)da
i1 (0, t)
0
a1

 a3
2 (a)e(a, t)da
+

m(S(t) S )
dL(t)
=
(1 R0 )i1 (0, t).
dt
S(t)

i1 (0)
i1 (0, t)
i1 (0, t)

(i) 0 = i1 (0)

+i2 (a1 )1 (a))i1 (a, t)da


 a2
+
(S 1 (a) + i2 (a) (m + 2 (a) + 2 (a))i2 (a)

dL(t)
dt

1 (a)i1 (a) 1

Proof. Using the boundary conditions of (7) and (22), we observe that

m(S(t) S )2
dL(t)
=
i1 (0, t) + R0 i1 (0, t)
dt
S(t)
 a1
+
(S 1 (a) + i1 (a) (m + 1 (a) + 1 (a))i1 (a)

From (24), we simplify

a1

(25)

(ii) 0 = i2 (a1 )


0. Every solution of system (7)

converges to A, where A is the largest invariant set in { dLdt(t) = 0}.


Notice that when R0 < 1, the equality { dL(t) = 0} holds only if S(t) = S

dt

and i1 (0, t) = 0. Using (9), we have i1 (a, t) = 0 for t > a in A. Moreover,


since the solution of (7) is non-negative, we deduce that i2 (a, t) = 0
and e(a, t) = 0 from the fth equation of (7). Hence we obtain that

A = E.
Let
When R0 = 1, the equality { dLdt(t) = 0} holds only if S(t) = S.
(S(t), i1 (a, t), i2 (a, t), e(a, t)) be the solution of (7) with the initial conditions in A. We have S(t) = S for any t from the invariance of A.
By the rst equation of (7), i.e., dSdt(t) = b mS(t) i1 (0, t), we have
b mS i1 (0, t) 0. Thus, we also obtain i1 (0, t) = 0. Therefore, we
have A = E by using the same method as R0 < 1. By the LaSalle
largest invariant set theorem, the disease-free equilibrium E is globally asymptotically stable if R0 1.
If R0 > 1, by continuity and (25), there exists a neighborhood of E
such that dLdt(t) > 0. Thus, there exists positive solutions of (7) which
This implies that E is unstable
are close to E and move away from E.
when R0 > 1.

a1

a1

i2 (a1 )
i2 (a1 , t) =
i2 (a1 , t)

(iii) 0 = e (a2 )


=

a2
a1

a2
a1

1 (a)i1 (a)da

i2 (a1 )
i2 (a1 , t)

i1 (a, t) i2 (a1 )
da,

i1 (a) i2 (a1 , t)

and

a1

1 (a)i1 (a, t)da

1 (a)i1 (a) 1

e (a2 )
e(a2 , t) =
e(a2 , t)

a2

a1

2 (a)i2 (a)da

e (a2 )
e(a2 , t)

2 (a)i2 (a, t)da


2 (a)i2 (a) 1

i2 (a, t) e (a2 )
da.
i2 (a) e(a2 , t)

Next we state and prove the global stability of the endemic equilibrium point.
Theorem 4.2. If R0 > 1, the unique endemic equilibrium E is globally
asymptotically stable.

M. Shen et al / Mathematical Biosciences 263 (2015) 3750



[B1 (r)1 (t r)fx B1 (r)1 (t r), i1 (t r)
 a1 t
+ i1 (tr)fx (B1 (r)1 (t r), i1 (tr))]dr+
i1 (t+r)

Proof. Let f (x, x ) = x x x ln xx for x, x > 0. Note that the function f (x, x ) has a unique minimum at x = x and f (x ) = 0. We dene
a Lyapunov functional W (t) = W1 (t) + W2 (t) + W3 (t) + W4 (t) with



W1 (t) = f S(t), S ,
W3 (t) =
W4 (t) =
where

i1 (a) =
i2 (a) =
e (a) =

a1
a3

1 (t + r)
f i10 (r)
, i1 (t + r) dr
1 (r)
 a1 t


i1 (t + r) m + 1 (t + r) + 1 (t + r)

a2


s
S 1 (s) + e (a2 )2 (s) e a (m+2 (l)+2 (l))dl ds,

a3

2 (s)e

s
a



= f B1 (t), i1 (0) +


(m+3 (l))dl ds.

i1 (a) = S 1 (a) i2 (a1 )1 (a) + m + 1 (a) + 1 (a) i1 (a),




i2 (a) = S 1 (a) e (a2 )2 (a) + m + 2 (a) + 2 (a) i2 (a),
e (a) = S 2 (a) + (m + 3 (a))e (a).

+i1 (0)i1 (0, t)i1 (0)

S
S
+i1 (0, t)
.
S(t)
S(t)


=

t
0

a1

i1 (t r)f B1 (r)1 (t r), i1 (t r) dr

a1 t

i1 (t + r)f i10 (r)


x
x
x x
xfx (x, x )+x fx (x, x ) = x 1
+x 1 ln x
x
x
x (x )2
x
= x x x ln = f (x, x ),
x
dW2 (t)
=
dt

i1 (0)f B1 (t) (0) +

(t r)
0


f B1 (r)1 (t r), i1 (t r) dr
 t

i1 (t r) m + 1 (t r) + 1 (t r)
0

, i1


i1

i1 (a)

t
0

1 (a)
1 (t a)

a1
0

i1 (a)f i1 (a, t), i1 (a) da

i1 (a)(m+ 1 (a) + 1 (a))f (B1 (ta)1 (a), i1 (a))da

a1
t

i1 (a) m + 1 (a) + 1 (a)

1 (a)
, i (a) da
1 (t a) 1


a1
0

(i1 (a) m + 1 (a) + 1 (a)



f i1 (a, t), i1 (a) da.

we have

1 (t + r)
, i (t + r) dr.
1 (r) 1

Using (26) and the following equality



i1 (a))f i1 (a, t), i1 (a) da
 a1




= f i1 (0, t), i1 (0)
S 1 (a) + S (A3 + A4 A5 )1 (a)

(a)
i1 (a)f i10 (a t) 1
, i (a) da
1 (a t) 1

a1

1 (a)
, i (a) da
1 (t a) 1



= f i1 (0, t), i1 (0) +

i1 (a)f B1 (t a)1 (a), i1 (a) da

i1 (a) m + 1 (a) + 1 (a) i10 (t a)


f i10 (t a)

When 0 t a1 , using (9), W2 (t) can be rewritten as


t

a1

i1 (a) m + 1 (a) + 1 (a) [B1 (t a)1 (a)



= f i1 (0, t), i1 (0) +


S
b mS(t) i1 (0, t)
S(t)


S
= 1
mS + i1 (0) mS(t) i1 (0, t)
S(t)

W2 (t) =

1 (a)
, i (a)
1 (t a) 1


1 (a)
, i1 (a) da
+ i1 (a)fx i10 (t a)
1 (t a)

It is clear that W (t) is continuously differentiable and nonnegative. We


calculate the derivatives of W1 (t), W2 (t), W3 (t) and W4 (t) separately.

i1 (a)f B1 (t a)1 (a), i1 (a) da


fx i10 (t a)
(26)

m(S(t) S
S(t)


f i10 (t a)

where Ai , i = 1, . . . , 5 are given by (17)(20). We calculate the derivatives of i1 (a), i2 (a) and e (a) as follows:



+ i1 (a)fx B1 (t a)1 (a), i1 (a) da +

i1 (0) = S (A1 + A2 A3 + A2 A4 A5 ) = 1,
i2 (a1 ) = S (A3 + A4 A5 ), e (a2 ) = S A5 ,



fx B1 (t a)1 (a), i1 (a)

dW1 (t)
=
dt

1 (t + r)
1 (t + r)
f i (r)
, i (t + r)
1 (r) x 10
1 (r) 1


1 (t + r)
+ i1 (t + r)fx i10 (r)
, i1 (t + r) dr
1 (r)

It is easy to verify that

i10 (r)


s
S 1 (s) + i2 (a1 )1 (s) e a (m+1 (l)+1 (l))dl ds,

i1 (a)f i1 (a, t), i1 (a) da,

a1
a

a1

e (a)f e(a, t), e (a) da,

a2



i2 (a)f i2 (a, t), i2 (a) da,

a2

W2 (t) =

43

Similarly, when t > a1 , we have

W2 (t) =


=

a1
0

i1 (a)f B1 (t a)1 (a), i1 (a) da

t
ta1

i1 (t r)f B1 (r)1 (t r), i1 (t r) da.

Thus,

i1 (0, t)
dW2 (t)
= i1 (0, t) i1 (0) i1 (0) ln
dt
i1 (0)
 a1


S 1 (a) + S (A3 + A4 A5 )1 (a)

i1 (a, t)
i1 (a, t) i1 (a) i1 (a) ln
da.
i1 (a)

44

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

Hence, for all t 0, the above equality holds. Similarly, using (10),
(11) and (26), we obtain

i2 (a1 , t)
dW3 (t)
= S (A3 + A4 A5 ) i2 (a1 , t) i2 (a1 ) i2 (a1 ) ln
dt
i2 (a1 )
 a2


S 1 (a) + S A5 2 (a)

a1

i2 (a, t) i2 (a) i2 (a) ln

By the last two equations of (7) and (22), it is easy to nd that



S (A3 + A4 A5 ) i2 (a1 , t) i2 (a1 ) S (A3 + A4 A5 )
 a1

1 (a) i1 (a, t) i1 (a) da = 0,


0

a2

a1

2 (a) i2 (a, t) i2 (a) da = 0.

dW1 (t) dW2 (t) dW3 (t) dW4 (t)


dW (t)
=
+
+
+
dt
dt
dt
dt
dt

2
m(S(t) S )
i1 (0, t)
S
S

=
i1 (0)
+ ln
+ i1 (0, t)
S(t)
S(t)
i1 (0)
S(t)

 a1
i
(
a,
t
)
1

S 1 (a) i1 (a, t) i1 (a) i1 (a) ln


da
i1 (a)
0
 a1
i (a, t)
1 (a)i1 (a) ln 1
da
+ S (A3 + A4 A5 )
i1 (a)
0
i2 (a1 , t)
i2 (a1 )

i2 (a, t)
S 1 (a) i2 (a, t) i2 (a) i2 (a) ln
da
i2 (a)
a1
 a2
i (a, t)
+ S A5
2 (a)i2 (a) ln 2
da
i2 (a)
a1
 a3
e(a2 , t)

S 2 (a)
S A5 e (a2 ) ln
e (a2 )
a2

e(a, t)
e(a, t) e (a) e (a) ln
da.
e (a)


From

i1 (0, t)

+ S A5
S A5
a3

a2

S 1 (a)i2 (a) ln


a2

a1
 a2
a1

i2 (a, t)
da
i2 (a)

2 (a)i2 (a) ln
2 (a)i2 (a) ln

i2 (a, t)
da
i2 (a)

e(a2 , t)
da
e (a2 )

e(a, t)
da
e (a)
 a1
+ S
1 (a)i1 (a)

S 2 (a)e (a) ln

m(S(t) S )2
=
S(t)
0

S
i1 (a, t)
i1 (0, t)
1
+ ln
ln
da
S(t)
i1 (a)
i1 (0)

 a1
i2 (a1 , t)
i (a, t)
1 (a)i1 (a) ln 1
ln
+ S (A3 +A4 A5 )
da
i1 (a)
i2 (a1 )
0

 a2
S
i2 (a, t)
i1 (0, t)
+ S
da
1 (a)i2 (a) 1
+ln
ln
S
(
t
)
i
(
a
)
i1 (0)
a1
2

 a2
e(a2 , t)
i (a, t)
2 (a)i2 (a) ln 2
ln
+ S A5
da
i2 (a)
e (a2 )
a1

 a3
e(a, t)
i1 (0, t)
S
2 (a)e (a) 1
+ln
ln
+ S
da.
S(t)
e (a)
i1 (0)
a2
Using Lemma 4.1(i)(iii), we obtain

Thus, we get

S (A3 + A4 A5 )i2 (a1 ) ln

a1

e(a2 , t)
dW4 (t)
= S A5 e(a2 , t) e (a2 ) e (a2 ) ln
dt
e (a2 )

 a3
e(a, t)

S 2 (a) e(a, t) e (a) e (a) ln


da.
e (a)
a2



S A5 e(a2 , t) e (a2 ) S A5

a2

i2 (a, t)
da,
i2 (a)

and

and


+

a2

 a1
 a2
S
S
1 (a)i1 (a, t)da +
1 (a)i2 (a, t)da
S(t)
0
a1

 a3
+
2 (a)e(a, t)da = 0
a2

and the last three equations of (22), we have

i1 (0, t)
m(S(t) S )2
S
dW (t)
i1 (0)
+ ln
=
+ i1 (0)
dt
S(t)
S(t)
i1 (0)
 a1
i1 (a, t)
+
S 1 (a)i1 (a) ln
da
i1 (a)
0
 a1
i (a, t)
+ S (A3 + A4 A5 )
1 (a)i1 (a) ln 1
da
i1 (a)
0
 a1
i (a , t)
S (A3 + A4 A5 )
1 (a)i1 (a) ln 2 1 da
i2 (a1 )
0


 a1
m(S(t) S )2
dW (t)
S
=
+ S
1 (a)i1 (a) 1
+1
dt
S(t)
S(t)
0

i1 (a, t)
i1 (0, t)
S(t) i1 (a, t) i1 (0)
+ ln
ln
da

S i1 (a) i1 (0, t)
i1 (a)
i1 (0)

 a1
i (a, t) i2 (a1 )
1 (a)i1 (a) 1 1
+ S (A3 + A4 A5 )
i1 (a) i2 (a1 , t)
0

i1 (a, t)
i2 (a1 , t)
+ ln
ln
da
i1 (a)
i2 (a1 )

 a2
S(t) i2 (a, t) i1 (0)
S
1 (a)i2 (a) 1
+1
+ S
S(t)
S i2 (a) i1 (0, t)
a1

 a2
i2 (a, t)
i1 (0, t)
+ ln
ln
2 (a)i2 (a)
da + S A5
i2 (a)
i1 (0)
a1

i2 (a, t) e (a2 )
i2 (a, t)
e(a2 , t)
1
+ ln
ln
da
i2 (a) e(a2 , t)
i2 (a)
e (a2 )

 a3
S(t) e(a, t) i1 (0)
S
2 (a)e (a) 1
+1
+ S
S(t)
S e (a) i1 (0, t)
a2

e(a, t)
i1 (0, t)
da
+ ln
ln
e (a)
i1 (0)

 a1
m(S(t) S )2
S
S
1 (a)i1 (a) g
=
S(t)
S(t)
0

S(t) i1 (a, t) i1 (0)


+g
da

S i1 (a) i1 (0, t)

 a1
i (a, t) i2 (a1 )
1 (a)i1 (a)g 1
S (A3 + A4 A5 )
da
i1 (a) i2 (a1 , t)
0

 a2
S
S(t) i2 (a, t) i1 (0)
1 (a)i2 (a) g
S
+g
da
S(t)
S i2 (a) i1 (0, t)
a1

 a2
i (a, t) e (a2 )
2 (a)i2 (a) 2
da
S A5
i2 (a) e(a2 , t)
a1

 a3
S
S(t) e(a, t) i1 (0)
2 (a)e (a) g
S
+g
da.
S(t)
S e (a) i1 (0, t)
a2
where g(x) = x 1 ln x 0 for all x > 0, and g(x) has a unique minimum at x = 1 and g(1) = 0. Thus, dWdt(t) 0 with equality holding if

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

and only if SS(t) = 1 and

transmission rate is estimated as

0.8578
0 V
.
=

b/m
1 + 0 V

i1 (a, t) i2 (a1 )
i2 (a, t) i1 (0)
i1 (a, t) i1 (0)
=
=

i1 (a) i1 (0, t)
i1 (a) i2 (a1 , t)
i2 (a) i1 (0, t)
=

i2 (a, t) e (a2 )
e(a, t) i1 (0)
=
= 1.

i2 (a) e(a2 , t)
e (a) i1 (0, t)

This second condition is equivalent to

i2 (a, t)
e(a, t)
i1 (0, t)
i1 (a, t)
=
=
=
,
i1 (a)
i2 (a)
e (a)
i1 (0)

t 0.

(27)

It remains to look for the largest invariant set M for which (27)
satises. In M, it is necessary to have S(t) = S for all t 0 and so we
have dSdt(t) = 0. Combining this with (27) and the rst equation of (22),
we get

i (0, t)
1 (a) 1
i (a)da
i1 (0) 1
0

 a2
 a3
i (0, t)
i (0, t)
+
1 (a) 1
i2 (a)da +
2 (a) 1
e (a)da
i1 (0)
i1 (0)
a1
a2
 a1
 a2
i1 (0, t)

S
1 (a)i1 (a)da +
1 (a)i2 (a)da
= b mS
i1 (0)
0
a1

0 = b mS S

a3
a2

a1

2 (a)e (a)da

i1 (0, t)
= (b mS ) 1
.
i1 (0)
b
we have i1 (0, t) = i (0). Thus, by (27), it follows
Since S = m
= S,
1
that i1 (a, t) = i1 (a), i2 (a, t) = i2 (a), e(a, t) = e (a) for all t 0 and we
conclude that M = {E }. Therefore, by the LaSalle largest invariant set
theorem, every solution of (7) converges to the endemic equilibrium
E . This shows that E is globally asymptotically stable whenever it
exists (i.e., R0 > 1).

5. Numerical results
Our main result on threshold dynamics shows that the global dynamics of the model (7) is determined by the basic reproduction number R0 , which involves both viral dynamic and epidemic parameters.
To examine the impact of within-host viral dynamics on betweenhost epidemiological dynamics, we investigate the variation in the
basic reproduction number R0 and the equilibrium level of total infected individuals by changing some parameters of interest. The initial
values of variables and baseline parameter values of the models (1)
and (7) are summarized in Table 1. The values of some parameters are
chosen as follows. We choose a1 = 90 days [5], a2 = 3139 days (i.e.,
8.6 years) [7], and a3 = 3832.5 days (i.e., 10.5 years) [36]. Thus, we
obtain the rate of progression to the asymptomatic stage is 1 = 1/90
per day. Similarly, we have the rate of progression to the AIDS stage
is 2 = 1/(a2 a1 ) = 1/3049 per day and the disease-induced death
rate at the AIDS stage is 3 = 1/(a3 a2 ) = 1/693.5 per day. The acute
stage is so short that we let the disease-induced death rate at this stage
be 1 (a) = 0 [6]. Note that the removal rate m consists of the natural
death rate m1 and the rate m2 at which people exit due to sexual
behavior change. Following [7] and [2], life expectancy and the mean
duration of sexual activity are 67 years and 32 years, respectively.
Thus, we choose m1 = 1/67 = 0.0149 per year, m2 = 1/32 per year,
and m = m1 + m2 = 0.0462 per year or 0.0462/365 per day. As c2 (a)
is assumed to be a decreasing function of the infection age in the AIDS
stage, we use the following simple function as the viral clearance rate

c(a) =

c1 ,

a [0, a2 ],

c2 (a) = c1

45

c1
(a a2 ),
a3 a2

a [a2 , a3 ].

Sun et al. [39] estimated the transmission coecient to be 0.8578


per year. With our notation, using the method in [4], the baseline

where V is given by (2). We choose one pair of baseline parameter values 0 = 105 and 0 = 1.848 109 which satisfy the above
equation. Note that highly active antiretroviral therapies (HAART)
are currently the most effective treatment regime for HIV patients
in terms of rapidly suppressing HIV viral load, which results in a
longer asymptomatic chronic stage. We assume that the diseaseinduced death rate 2 (a) is linearly dependent on the viral load,
that is, 2 (a) = 2 V1 (a), a [a1 , a2 ], as used in [23,26,27]. According to the estimated disease-related death rate 0.172 per year in [7],
we have 2 V1 (a) = 0.172, a [a1 , a2 ]. This gives a baseline estimate
2 = 1.8 106 .
We illustrate the long-term behavior of the system (7) in Fig. 1
when R0 > 1 and R0 < 1. It is shown that the system converges to
the endemic steady state when R0 > 1 (as shown in Fig. 1(a) and
(b)), while the disease is predicted to die out when R0 < 1 (shown
in Fig. 1(c) and (d)). This agrees with the conclusion in Theorems 3.1
and 4.2.
To explore the effect of within-host dynamics on the basic reproduction number R0 and the total number of infected individuals, we consider the case in which all the infected individuals at the
asymptomatic stage are given a combination of reverse transcriptase inhibitors and protease inhibitors. The effectiveness of these
inhibitors is RT and PI , respectively. Because of the therapy, the
duration from infection to the AIDS stage after treatment becomes
longer and is denoted as a 2 (a 2 a2 ). These treated infectives proceed to the AIDS stage with continuous therapy and survive to time
a 3 , where a 3 = a 2 + a3 a2 (which implicitly assumes that the duration of the AIDS stage remains unchanged after combination therapy).
As for the corresponding within-host model, we substitute k(1 RT )
and p(1 PI ) for k and p of the model (1), respectively, and keep
the other parameters xed. Denote the concentration of infectious
virus particles at the asymptomatic stage and the AIDS stage as VI1 (a)
and VI2 (a), respectively. After the treatment, the basic reproduction
number R0 in (21) becomes

R0 =

b
(A1 + A2 A3 + A2 A4 A5 ),
m

where A1 , A2 are given by (17) and (18), and

A3 =

A5 =

a 2
a1
a 3
a 2

0 VI1 (a)
2 (a)da, A4 =
1 + 0 VI1 (a)
0 VI2 (a) (m+3 )(aa 2 )
e
da.
1 + 0 VI2 (a)

a 2

a1

2 2 (a)da,

And the equilibrium level of the total infected people, denoted by I ,


gives

I = I1 + I2 + A =


0

a1

i1 (a)da +

a 2

a1

i2 (a)da +

a 3
a 2

e (a)da.

Here i1 (a), i2 (a) and e (a) are given by (23) where R0 and A4 are be
replaced with R0 and A 4 , respectively.
Let n be the average prolonged lifespan of the asymptomatic stage
due to ART, i.e., n = a 2 a2 . Fig. 2 shows that for any given treatment
ecacy both the basic reproduction number R 0 and the equilibrium
I of the total infected individuals increase as the prolonged lifespan n increases. Fig. 2(a)(d) shows that increasing any treatment
ecacy (either RT or PI ) leads to the decline of the basic reproduction number R 0 (hence the HIV new infection at the population
level). Specically, R0 declines very slowly when increasing the effectiveness of RT inhibitors RT initially until a critical level ( RT ) above
which R 0 begins to decline quickly. This indicates that a lower treatment ecacy is not enough to effectively control the HIV infection.

46

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

(a)

(b)

10000

1400
I1(t)
1200

I2(t)

8000

A(t)
Populations

Population

1000
6000
S(t)
4000

800
600

I (t)
2

400
2000
200
0

10

50

100

A(t)

I (t)
0 1
10

150

50

time t (years)

100

150

time t (years)

(c)

(d)
600

10000

I1(t)
S(t)

I (t)

500

8000

A(t)
Populations

Population

400
6000

4000

300
I2(t)

200
2000

100

10

50

100
time t (years)

150

A(t)

I (t)
0 1
10

50

100

150

time t (years)

Fig. 1. The dynamic behaviors of the model Eq. (7) are depicted when R0 > 1 and R0 < 1 with the baseline parameter values given in Table 1. (a) and (b) When 0 = 1.848 109 ,
b
= 5626.1. The curves of I1 (t), I2 (t) and A(t) indicate
we obtain R0 = 1.686. The disease becomes endemic, and the susceptible population S(t) converges to the steady state S = mR
a
a
a 0
that the disease peaks in about 8 years. The steady states are I1 = 0 1 i1 (a)da = 27.7, I2 = a12 i2 (a)da = 312.1, and A = a23 e (a)da = 43.3. (c) and (d) 0 = 1.07 109 and the
other parameter values are the same as (a) and (b). In this case, we have R0 = 0.9762. The disease becomes extinct and the susceptible population S(t) converges to the steady state
S = mb = 9485.6.

Moreover, it is interesting to note that Fig. 2(e) shows, for a relatively long prolonged lifespan (say, n = 10), increasing RT initially
increases the equilibrium level of total infected individuals I . However, I quickly declines when RT exceeds a critical level ( RT ), as
R 0 does. This can be explained as follows. Treatment with low levels
of ecacy RT barely decreases HIV new infection, but reduces the
within-host viral load which results in a decreased disease-induced
mortality rate 2 (a) and a longer lifespan of infected people. Thus,
relatively unchanged new infections and less death due to a declining
disease-induced death rate result in an increase in the equilibrium
level of the overall infected individuals. However, when the treatment ecacy RT exceeds a critical level RT , rapid decrease in R0 implies a quick decline in HIV new infection, which, together with less
death of infected individuals due to a smaller disease-induced death
rate, leads to a decline in the equilibrium level of the overall infected
individuals.
Comparing Fig. 2(f) with (e) implies that, for a given prolonged
lifespan (n = 6), additionally introducing a low ecacy of PI inhibitors

does not decrease the equilibrium level of total infected individuals


I (for example, I > 600 for PI = 0.2 and n = 6 shown in Fig. 2(f),
while I < 600 for PI = 0 and n = 6 shown in Fig. 2(e)). This is because new infection almost does not decrease (see the green color
curve in Fig. 2(a) and (b)), but within-host viral loads are reduced
because of an additional introduction of PI inhibitors. This again induces a decreased disease-induced mortality rate 2 (a). Thus, this
long-term accumulation inevitably increases the equilibrium level of
total infected individuals I . With an increasing eciency of PI inhibitors PI , the equilibrium level of total infected people I keeps
almost unchanged before a quick decline, as shown in Fig. 2(g), due to
less new infection balancing a decreased disease-related death rate.
It follows from Fig. 2(d) and (h) that if the eciency of drugs is high,
both new infection and the total infected people at equilibrium I decrease with increasing RT no matter how long the prolonged lifespan
is. This means treatment is benecial to the control of the disease only
with a high level of drug ecacy. This nding suggests that treatment
may induce more new infection or a higher equilibrium level of total

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

(a)

(b)

47

(c)

(d)

2.2

2.2

1.8

1.8

1.8

1.6

1.6

1.6

n=0
n=3
n=6
n=10

1
0.7

0.5
RT (PI=0)

n=0
n=3
n=6
n=10

1
0.7

1.8
1.6

(e)

R0

R0

R0

2.2

R0

2.2

0.5
RT (PI=0.2)

0.7

n=0
n=3
n=6
n=10
0

(f)

0.5
RT (PI=0.4)

0.3

n=0
n=3
n=6
n=10
0

(g)
800

800

600

600

600

600

n=0
n=3
n=6
n=10

200

RT

0.5
( =0)

400
n=0
n=3
n=6
n=10

200

PI

I*

I*

800

400

400
n=0
n=3
n=6
n=10

200

0.5
RT (PI=0.2)

(h)

800

400

0.5
RT (PI=0.6)

0.5
RT (PI=0.4)

n=0
n=3
n=6
n=10

200

0.5
RT (PI=0.6)

Fig. 2. The basic reproduction number R 0 and total infected people at the equilibrium I versus the eciency of RT inhibitor RT . The extended lifespan is chosen to be n = 0, 3, 6, 10
years. (a)(d) R 0 is plotted against RT with PI = 0, 0.2, 0.4, 0.6, respectively. (e)(h) I is plotted against RT with PI = 0, 0.2, 0.4, 0.6, respectively. All the other parameters
are shown in Table 1. (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)

infected individuals in the long run, depending on the drug ecacy


and a prolonged lifespan.
Note that our main theoretical results are obtained when the
disease-induced death rates and the rates of progression to asymptomatic and AIDS stages are assumed to be general age-dependent
functions in the model. In the above simulations, we chose these
rates to be constant. It would be interesting to compare the above
prediction with the simulation assuming age-dependent diseaseinduced death rates and the rates of progression to asymptomatic
and AIDS stages. For example, we choose 1 (a), 2 (a) and 3 (a) as
age-dependent functions as follows used in [9],

1 (a) = 1 (1 e0.1a ), a [0, a1 ],


2 (a) = 2 (1 e0.005(aa1 )), a [a1 , a2 ],
3 (a) = 3 (1 e0.01(aa2 )), a [a2 , a3 ],

(28)

where 1 , 2 and
3 have the same values as constant values of 1 , 2
and 3 in Table 1, respectively, i.e., 1 = 1/90, 2 = 1/3049,
3 =
1/693.5. Obviously these time-varying rates are smaller than previous constant rates. With the age-dependent rates, we plotted the
basic reproduction number and the total infected population again in
Fig. 3. The dynamics are quite similar to those shown in Fig. 2. However, the magnitudes in each subgure become smaller. This suggests
that as the transition rate to the next stage of HIV progression becomes
smaller, both the basic reproduction number and the total infected
population become smaller.
In order to understand which parameter most impacts the basic
reproductive number, we perform a sensitivity analysis using the

Latin Hypercube Sampling (LHS) technique and calculate the partial


rank correlation coecients (PRCCs) [40,41] for the input parameters
listed in Table 2 against output variable R0 . The magnitude of these
PRCCs shows the sensitivity of these parameters and the sign of the
PRCCs indicates the positive or negative relationship between the
inputs and outputs. Because of limited information on the distribution
for each parameter, we choose a uniform distribution as in [41] for
all input parameters with ranges listed in Table 2. According to the
ranges of the transmission coecient [0.1, 1] and the disease-related
death rate [0.01, 1] in [39], we deduce the corresponding ranges of
0 , 0 and 2 . They are listed in Table 2.
We list the PRCCs in Table 2 and nd that there exists a close
correlation between R0 and both within-host and between-host parameters. It follows from Fig. 4 that among the viral dynamical parameters, the rst two parameters with the most signicant impact
on R0 are the effectiveness of inhibitors PI and the rate of clearance
of virions c1 . It is reasonable that the above two parameters play a
very important role in decreasing HIV new infections. That is, a larger
eciency of inhibitors and a larger clearance rate of virions denitely
result in a lower R0 . Of the epidemiological parameters, the recruitment rate b, the saturated coecient 0 and the removal rate m2 due
to behavior changes are the most critical in affecting R 0 . Note that a
lower recruitment rate and a higher removal rate result in a decline
of the basic reproductive number R0 . This suggests that a possible
control strategy is to educate individuals to give up the high-risk sexual behaviors. It is interesting that treatment strategy, on one hand,
effectively reduces the viral load and lowers the transmission rate
which makes R 0 decrease, but on the other hand, extends the lifespan

48

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

(a)

(b)

(c)

1.8

1.8

1.8

1.6

1.6

1.6

1.4

1.4

1.4

(d)
2

1.6

1.2

1.2

n=0
n=3
n=6
n=10

1
0.8
0

0.8

0.5
RT (PI=0)

1.2

n=0
n=3
n=6
n=10

0.5
RT (PI=0.2)

(e)

R0

R0

R0

R0

1.4

1
0.8
1

n=0
n=3
n=6
n=10
0

0.5

0.5
RT (PI=0.4)

(f)

n=0
n=3
n=6
n=10

(g)

600

600

600

500

500

500

500

400

400

400

400

300

300

300

300

n=0
n=3
n=6
n=10

100
0

RT

0.5
( =0)

n=0
n=3
n=6
n=10

200
100
1

PI

RT

0.5
( =0.2)

n=0
n=3
n=6
n=10

200
100
0

PI

RT

(h)

600

200

0.5
RT (PI=0.6)

n=0
n=3
n=6
n=10

200
100

0.5
( =0.4)

PI

RT

0.5
( =0.6)

PI

Fig. 3. The basic reproduction number R 0 and total infected people at the equilibrium I versus the eciency of RT inhibitor RT . All the conditions are the same as Fig. 2 except
that 1 (a), 2 (a) and 3 (a) are chosen as (28).
Table 2
PRCC values for R 0 .
Input parameters

Distributions

Source

R0
PRCC

p-value

RT
PI

U (0.01, 1)
U (0.01, 1)
U (0.1, 10)
U (3139, 6789)
U (0.01, 5)
U ( 0.02
, 0.2 )
365 365
U (107 , 105 )
U (106 , 104 )
U (2 1011 , 1.2 108 )

[9]
[9]
[37]
[36]

[3]
[39]

[39]

0.1779
0.3632
0.2000
0.0556
0.5437
0.3985
0.0832
0.2518
0.5598

0
0
0
0.0806
0
0
0.0089
0
0

c1
a 2
b
m2

2
0
0

of the infected individuals (a larger a 2 ) which induces an increase


in I .
6. Discussion
In this study we have formulated a multi-scale nested model to
describe HIV infection by connecting two separate levels, i.e., the viral dynamics at the individual level and the transmission dynamics at
the population level. Since the progression of an HIV infection often
advances through three typical stages, corresponding to drastically
different CD4+ T cell and viral levels, a simple way to track the viral
load precisely in infected hosts is the infection age a, which begins at
a = 0 upon infection. Considering this, we propose a staged progression PDE model that includes continuous age-structure for infected
people at three distinct stages. This is more realistic than ordinary

differential equations (ODEs). Furthermore, the transmission rates at


different stages are treated as a saturated function of viral load, which
is more reasonable compared with a linear relationship [2022]. Thus,
the epidemiological parameters 1 (a) and 2 (a) are closely tied to
the state of the infection within hosts. This allows the immunological
dynamics to have its own time-scale and unies the time-scales at
within-host and between-host levels. Hence we can keep track of the
infection status of infected hosts at different ages of infection.
In order to determine the asymptotic behavior of solutions, we
have examined the global stability of the coupled system by constructing appropriate Lyapunov functionals [11,2830]. We obtain
that if R0 1, the disease-free equilibrium is globally asymptotically
stable (see Fig. 1(c) and (d)). In this case, the virus will go extinct
eventually. When R0 > 1, the disease persists and the unique endemic equilibrium is globally asymptotically stable (see Fig. 1(a) and

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

References

*significant
(pvalue<0.01)

0.8
0.6

PRC C s for R0

0.4
0.2
0
0.2

0.4

0.6
0.8
1

49

RT

PI

a2

Fig. 4. PRCC results illustrating the dependence of R0 on both immunological parameters and epidemiological parameters. Sample size is set to 1000. (*) denotes PRCCs are
signicant (p-value <0.01). Parameter values and ranges are shown in Tables 1 and 2.

(b)). This means that the virus infection establishes within a host and
the disease prevails among the population.
Since the transmission parameters are likely to vary over the progression of an infection due to coincident interaction between cells
and virus within a host, we studied the variations in R0 with respect
to both immunological parameters and epidemiological parameters
(see Fig. 4). This is an important advantage of using the nested model.
It overcomes the traditional shortcoming that regards the transmission rate as an independent identical constant [2,4,6,7,20,21] in all
infected individuals. It follows from our sensitivity analysis (Fig. 4)
that the basic reproduction number is sensitive to the effectiveness
of inhibitors, the clearance rate of virions, the recruitment rate of
susceptible population, the saturated coecients and the removal
rate due to changes in sexual behavior. In particular, improving drug
ecacy and reducing the sexually active duration greatly decrease
the basic reproduction number and hence result in a decline in new
infection.
Combination therapy can decrease the viral load and thus the infectiousness of each infected host and the disease-induced mortality
rate. On the other hand, since the infected people live longer, it may
result in a higher chance to infect others. This indicates a complicated
effect of the treatment on HIV epidemics. Our numerical studies indicate that the treatment may be harmful to HIV infection control in
some cases (Fig. 2(e) and (f)) as more infected people appear. However, in such cases the disease can still be controlled if the effectiveness of treatment is very high (Fig. 2(h)). These results are based
on our assumption that the within-host viral load determines the
disease-related death rate at the asymptomatic stage, i.e., decreasing within-host viral load results in a decrease in the disease-related
mortality rate. The functional form of this relationship (such as the
saturated relationship used in our paper) does not affect our results.
These conclusions rene our understanding of the disease and may
facilitate the extraction of more information from both levels (see
Figs. 2 and 4) that cannot be obtained from the pure between- [17]
or within-host [811] models.
Acknowledgments
The authors are supported by the National Mega-project of Science Research no. 2008ZX10001-003, by the National Natural Science Foundation of China (11171268 (Y.X.)), by the Fundamental Research Funds for the Central Universities (08143042 (Y.X.)), and by
NSF grants DMS-1122290 and DMS-1349939 (L.R.).

[1] H.R. Thieme, C. Castillo-Chavez, How may infection-age-dependent infectivity


affect the dynamics of hiv/aids?, SIAM J. Appl. Math. 53 (1993) 1447.
[2] A.R. McLean, S.M. Blower, Imperfect vaccines and herd immunity to HIV, Proc. R.
Soc. London, Ser. B 253 (1993) 9.
[3] J.M. Hyman, J. Li, E.A. Stanley, The differential infectivity and staged progression
models for the transmission of HIV, Math. Biosci. 208 (1999) 227.
[4] A.B. Gumel, C.C. McCluskey, P. van den Driessche, Mathematical study of a
staged progression HIV model with imperfect vaccine, Bull. Math. Biol. 68 (2006)
2105.
[5] Y.C. Zhou, Y.M. Shao, Y.H. Ruan, J.Q. Xu, Z. Ma, C.L. Mei, J.H. Wu, Modeling and
prediction of HIV in china: transmission rates structured by infection ages, Math.
Biosci. Eng. 5 (2008) 403.
[6] X.X. Xu, Y.N. Xiao, N. Wang, Modeling sexual transmission of HIV/AIDS in Jiangsu
province, China, Math. Methods Appl. Sci. 36 (2013) 234.
[7] Y.N. Xiao, S.Y. Tang, Y.C. Zhou, J.S. Robert, J.H. Wu, N. Wang, Predicting an HIV/AIDS
epidemic and measuring the effect on it of population mobility in mainland China,
J. Theor. Biol. 317 (2013) 271.
[8] A.S. Perelson, P.W. Nelson, Mathematical analysis of HIV-1 dynamics in vivo, SIAM
Rev. 41 (1999) 3.
[9] L.B. Rong, Z.L. Feng, A.S. Perelson, Mathematical analysis of age-structured HIV-1
dynamics with combination antiretroviral therapy, SIAM J. Appl. Math. 67 (2007)
731.
[10] S.Y. Tang, Y.N. Xiao, N. Wang, H.L. Wu, Piecewise HIV virus dynamic model with
CD4+ T cell count-guided therapy: I, J. Theor. Biol. 308 (2012) 123.
[11] G. Huang, X.N. Liu, Y. Takeuchi, Lyapunov functions and global stability for agestructured HIV infection model, SIAM J. Appl. Math. 72 (2012) 25.
[12] Y. Xiao, H. Miao, S. Tang, H. Wu, Modeling antiretroviral drug responses for HIV-1
infected patients using differential equation models, Adv. Drug Delivery Rev. 65
(2013) 940.
[13] T.C. Quinn, M.J. Wawer, N. Sewankambo, et al. Viral load and heterosexual transmission of human immunodeciency virus type 1. Rakai Project Study Group, N.
Engl. J. Med. 342 (2000) 921.
[14] B.L. Rapatski, F. Suppe, J.A. Yorke, Hiv epidemics driven by late disease stage
transmission, J. Acquired Immune. Dec. Syndr. 38 (2005) 241.
[15] M.J. Wawer, R.H. Gray, N.K. Sewankambo, et al. Rates of HIV-1 transmission per
coital act, by stage of HIV-1 infection, in Rakai, Uganda, J. Infect. Dis. 191 (2005)
1403.
[16] C. Fraser, T.D. Hollingsworth, R. Chapman, F. de Wolf, W.P. Hanage, Variation in
HIV set-point viral load: epidemiological analysis and an evolutionary hypothesis,
Proc. Natl. Acad. Sci. U.S.A. 104 (2007) 17441.
[17] D.P. Wilson, M.G. Law, A.E. Grulich, D.A. Cooper, J.M. Kaldor, Relation
between HIV viral load and infectiousness: a model-based analysis, Lancet 372
(2008) 314.
[18] T.D. Hollingsworth, R.M. Anderson, C. Fraser, HIV-1 transmission, by stage of
infection, J. Infect. Dis. 198 (2008) 687.
[19] L.J. Abu-Raddad, R.V. Barnabas, H. Janes, H.A. Weiss, J.G. Kublin, I.M.L. JR, JR,
J.N. Wasserheit, Have the explosive HIV epidemics in sub-Saharan Africa been
driven by higher community viral load? AIDS 27 (2013) 981.
[20] M.A. Gilchrist, D. Coombs, Evolution of virulence: interdependence, constrains,
and selection using nested models, Theor. Popul. Biol. 69 (2006) 145.
[21] D. Coombs, M.A. Gilchrist, C.L. Ball, Evaluating the importance of within- and
between-host selection pressures on the evolution of chronic pathogens, Theor.
Popul. Biol. 72 (2007) 576.
[22] Z.L. Feng, J.V. Hernandez, B.T. Santos, M.C.A. Leite, A model for coupling withinhost and between-host dynamics in an infectious disease, Nonlinear Dyn. 68
(2012) 401.
[23] M.A. Gilchrist, A. Sasaki, Modeling host-parasite coevolution: a nested approach
based on mechanistic models, J. Theor. Biol. 218 (2002) 289.
[24] A. Sasaki, Y. Iwasa, Optimal growth schedule of pathogens within a host: switching
between lytic and latent cycles, Theor. Popul. Biol. 39 (1991) 201.
[25] N. Mideo, S. Alizon, T. Day, Linking within- and between-host dynamics in the
evolutionary epidemiology of infectious diseases, Trends Ecol. Evol. 23 (2008)
511.
[26] M. Martcheva, X.Z. Li, Linking immunological and epidemiological dynamics of
HIV: the case of super-infection, J. Biol. Dyn. 7 (2013) 161.
[27] E. Numfor, S. Bhattacharya, S. Lenhart, M. Martcheva, Optimal control of nested
within-host and between-host model, Math. Modell. Nat. Phenom. 7 (2014) 32.
[28] P. Magal, C.C. McCluskey, G.F. Webb, Lyapunov functional and global asymptotic
stability for an infection-age model, Appl. Anal. 89 (2010) 1109.
[29] C.C. McCluskey, Global stability for an SEI epidemiological model with continuous
age-structure in the exposed and infectious classes, Math. Biosci. Eng. 9 (2012)
819.
[30] F. Brauer, Z.S. Shuai, P. van den Driessche, Dynamics of an age-of-infection cholera
model, Math. Biosci. Eng. 10 (2013) 1335.
[31] P. De Leenheer, H.L. Smith, Virus dynamics: a global analysis, SIAM J. Appl. Math.
63 (2003) 1313.
[32] M.E. Gurtin, R.C. MacCamy, Nonlinear age-dependent population dynamics, Arch.
Ration. Mech. Anal. 54 (1974) 281.
[33] G.F. Webb, Theory of Nonlinear Age-Dependent Population Dynamics, Marcel
Dekker, New York, 1985.
[34] J.K. Hale, P. Waltman, Persistence in nite dimensional systems, SIAM J. Math.
Anal. 20 (1989) 388.
[35] J.K. Hale, Asymptotic Behavior of Dissipative Systems, Mathematical Surveys and
Monographs, 25, American Mathematical Society, Providence, 1988.

50

M. Shen et al / Mathematical Biosciences 263 (2015) 3750

[36] T.B. Hallett, S. Gregson, O. Mugurungi, E. Gonese, G.P. Garnett, Assessing evidence
for behaviour change affecting the course of HIV epidemics: a new mathematical
modelling approach and application to data from Zimbabwe, Epidemics 1 (2009)
108.
[37] Y.P. Yang, Y.N. Xiao, N. Wang, J.H. Wu, Optimal control of drug therapy: melding
pharmacokinetics with viral dynamics, Biosystems 107 (2012) 174.
[38] M.A. Stafford, L. Corey, Y. Cao, E.S. Daar, D.D. Ho, A.S. Perelson, Modeling
plasma virus concentration during primary HIV infection, J. Theor. Biol. 203 (2000)
285.

[39] X.D. Sun, Y.N. Xiao, Z.H. Peng, N. Wang, Modelling hiv/aids epidemic
among men who have sex with men in china, BioMed Res. Int. (2013).
doi:10.1155/2013/413260.
[40] S.M. Blower, H. Dowlatabadi, Sensitivity and uncertainty analysis of complexmodels of disease transmission an HIV model, as an example, Int. Stat. Rev. 62
(1994) 229.
[41] S. Marino, I.B. Hogue, C.J. Ray, D.E. Kirschner, A methodology for performing global
uncertainty and sensitivity analysis in systems biology, J. Theor. Biol. 254 (2008)
178.

S-ar putea să vă placă și