Global Dynamics of Infectious Disease with Arbitrary Distributed Infectious Period on Complex Networks

Most of the current epidemic models assume that the infectious period follows an exponential distribution. However, due to individual heterogeneity and epidemic diversity, these models fail to describe the distribution of infectious periods precisely. We establish a SIS epidemic model with multistaged progression of infectious periods on complex networks, which can be used to characterize arbitrary distributions of infectious periods of the individuals. By using mathematical analysis, the basic reproduction number R 0 for the model is derived. We verify that the R 0 depends on the average distributions of infection periods for different types of infective individuals, which extend the general theory obtained from the single infectious period epidemic models. It is proved that if R 0 < 1, then the disease-free equilibrium is globally asymptotically stable; otherwise the unique endemic equilibrium exists such that it is globally asymptotically attractive. Finally numerical simulations hold for the validity of our theoretical results is given.


Introduction
The infectious period of an infective individual means the period during which an infected person has a probability of transmitting the virus to any susceptible host or vector they contact.Note that the infectious period may be associated with the fitness of persons.The influence degrees of infection and rates of disease transmission are varied for individuals with different infectious periods.Every year, some emerging infectious diseases with unknown infectious period are seriously threatening the health of people.There is no doubt that the deficiency of the infectious period's knowledge results in the difficulty of controlling epidemic.Then, in order to obtain the date of the infectious period of these epidemics in medicine, a large amount of statistics data is necessary.However, it is hard to get the date in the early stage of the disease.Therefore applying mathematical methods to research the effects of infectious period distribution on the infectious diseases spread is significative.
As the SIS compartment model was first proposed by Kermack and McKendrick in 1932 [1], thousands of scientists successively started to study the epidemic propagation by mathematic models [2][3][4].In most of their models, infected compartment contains all infected individuals and the proportion of infected individuals who transit into the next state per unit time is a constant .Wearing et al. [5] pointed out that the assumption of exponentially distributed infectious periods always results in underestimating the basic reproductive ratio of an infection from outbreak data.According to the staged progression features of HIV or TB, Lloyd [6] applied gamma distribution to describe the infectious period distribution.However, the distributions of the infectious period of a lot of infectious diseases in the real world may not satisfy exponent or gamma distribution.Then, Feng et al. [7] used integral-differential equations to study the nonexponential distribution of the infectious period.The homogeneous mixing models, they considered, ignore the heterogeneity of contacts of individuals.
The network origins from the well-known six degrees of separation theory.The small-world (SW) property is the most popular feature in complex network theory [8].The SW networks constitute a mathematical model for social networks that show two types.The first type can be called exponential networks since there is the probability of finding a node with degree  different from the average degree ⟨⟩ which decays exponentially fast for large .The second kind of networks comprises those referred to as scale-free (SF) networks.For these networks, the probability that a given node is connected to  other nodes follows a power law of the form () ∼  −V , with the remarkable feature that 2 < V ⩽ 3 for most real world networks.
With the development of networks research, there have been some researchers studying infectious disease models on networks for decades, such as SIS model proposed by Pastor-Satorras and Vespignani in 2001 [9], SIR model established by Yang et al. in 2007 [10], and SI pattern model introduced by Barthélemy et al. in 2004 [11].Hence studying the epidemic model with infectious period distribution on networks is necessary and meaningful.However, there is little literature about the infectious period distribution problems based on networks.Zhang et al. in 2011 proposed a susceptibleinfected-susceptible staged progression and different infectivity models on different complex networks [12], where the infectious period follows a gamma distribution.Zager and Verghese in 2009 established a discrete differential equation to present an arbitrarily distributed infectious period [13].Moreover, they studied epidemic thresholds for infections on uncertain networks.However, the former researchers did not analyze the stability of equilibrium theoretically.Therefore, we build continuous-time ordinary differential equations to study an arbitrarily distributed infectious period epidemic model on networks.We extend the scope of previous works in this area to include dynamic analysis results mathematically.
The rest of the paper is organized as follows.In Section 2, we establish an SIS model with an arbitrary distribution of infectious period on complex networks.We compute the basic reproduction number and analyze the globally asymptotic stability of the disease-free equilibrium and the global dynamics of the endemic equilibrium in Section 3. In Section 4, we perform numerical simulations to verify the above theories.Finally, a brief conclusion and discussion will be given in Section 5.

The Model
One feature of some diseases is that different patients may have different symptoms.The feature had appeared in some patients of some SIR/SI diseases, such as TB/HIV studied by Guo and Li in 2006 [14].However, the other feature that different groups may have different infected stage processes is ignored.We propose a modified staged progression model to capture the second feature above.In our model, the infectious period distributions of different groups follow different gamma distributions.The linear combination of different gamma distributions can be transformed into normal distribution, chi-square distribution, exponential distribution, Erlang distribution, or beta distribution [15].So our model can be transformed into any distributions.
We classify the population as infected  (,) and susceptible .Compartment  (,) contains those individuals whose infectious period has  stages and which are now in the th stage ( ⩽ ).Susceptible individuals enter into the first stage  (,1) after being infected and then gradually progress from this stage to stage  (,) ( = 1, 2, . . .,  ⩽ ).The infected individuals in the  (,) stage will be susceptible again.
In contrast to classical compartment models, we consider the whole population and their contacts on networks.Each individual in the community can be regarded as a vertex in the network, and each contact between two individuals is represented as an edge (line) connecting these vertices.The number of edges emanating from a vertex, that is, the number of contacts a person has, is called the degree of the vertex.Therefore, we assume that the population is divided into  distinct groups of sizes   ( = 1, 2, . . .,  → ∞) such that each individual in group  has exactly  contacts per day.If the whole population size is  ( =  1 +  2 + ⋅ ⋅ ⋅ +   ), then the probability that a uniformly chosen individual has  contacts is () =   /, which is called the degree distribution of the network.The average degree is ⟨⟩ = ∑  =1 ().
Let   () denote the number of susceptible nodes of degree  at time .Let  (,) () be the number of infected nodes whose infectious period has  stages and which are now in the th stage of their infection ( ⩽ ) of degree  at time .The transmission sketch is shown in Figure 1.
We make the following basic assumptions about the infectious disease models.
(1) Here, we will not incorporate the possibility of individual removal due to birth and death or acquired immunization.That is to say, the total population The stages in infectious period can be given by the discrete random variable .The range of values of  need not to be finite, but for ease of presentation we assume that  can only take values from 1 to .The probability of infected individuals with an infectious period of exactly  stages is   .
(3) The mutants of virus can cause the same individual to suffer different infected stage progression processes if he or she is infected again.(5) In a network with no assortative (i.e., disassortative) then the conditional probability (  | ) that a given vertex with degree  is linked to a vertex with degree   by one edge is proportional to (), which is independent of its own vertex degree  [9], and hence we have (  | ) =   (  )/⟨⟩.The expectation that any given edge points to an infected vertex becomes On the basis of the above assumption, the model of ((+1)/2+1) ordinary differential equations with  ( ⩾ 1) the maximum degree and  ( ⩾ 1) the maximum infection stages in infectious period is as follows: where  = 1, . . ., ,  = 1, . . ., , and 2 ⩽  ⩽ .For system (1), since the total population is constant, we can only consider the infected compartments below.For a given degree distribution (), the number of the nodes which have the same degree ,   =   +   = (), is definite, where . The relative densities of susceptible and infected nodes with an infectious period of  stages and which are now in the th stage of degree  at time  are denoted by   () =   ()/  () and  (,) () =  (,) ()/  (), respectively.Without loss of generality, we set  = 1, since it only affects the definition of the time scale of the epidemic transmission.
It is easy to verify that system (2) has a unique diseasefree equilibrium  0 = N ⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ (0, 0, . . ., 0).We note that only compartments  (,) are involved in the calculation of  0 .In the disease-free state  0 , the rate of appearance of new infections  and the rate of transfer of individuals out of the compartments  are given by where where where  * is the M × M matrices in which  * is a block diagonal matrix with  blocks and  *  are  ×  matrices with entries of 1 on the diagonal and −1 on the first subdiagonal.
Using the concepts of next-generation matrix [16], the reproduction number is given by  0 = ( −1 ), that is, the spectral radius of the matrix  −1 .We first represent the inverse of  by the following matrix: and since  * is block diagonal with blocks  *  , its inverse will be block diagonal with blocks  −1 *  which is  ×  lower triangular matrices with entries of 1.
Setting  =  −1 , we have where   (1 ⩽  ⩽ , 1 ⩽  ⩽ ) are M × M matrices and where where   = (,  − 1,  − 2, . . ., 1) 1×M .Now we are ready to compute the eigenvalues of the matrix  =  −1 .Thus, we obtain that the basic reproduction number  0 which is the largest modulus of the roots of the characteristic equation below To simplify and compute (11), we have Therefore, we obtain the reproduction number If  0 > 1, we can derive the critical transmission rate from (13), which is Because the degree distribution of scale-free network is () ∼  −V , with 2 < V ⩽ 3 in most cases, for which ⟨ 2 ⟩ → +∞, when the size of network is sufficiently large, then inequality ( 14) is always satisfied.In other words, the multistaged progression model will prevail on sufficiently large heterogenous networks more easily.

Global Stability of Disease-Free Equilibrium.
In order to study the global stability of the disease-free equilibrium  0 , we first give the following lemma, which guarantees that the densities of each infected class cannot become negative and the sum of the densities of infective individuals with the same degree cannot be greater than unity.
Denote () = max 1⩽⩽ Re(  ), where   for  = 1, . . ., N are the eigenvalues of , and Re represents the real part of the eigenvalues.
To obtain the global stability of the disease-free equilibrium  0 , we need the following lemma.

Global Attractivity of Endemic Equilibrium.
In the following, we will compute the value of the unique endemic equilibrium  * of the system (1), and we will use the method in [17] to ascertain global attractivity of the nonzero solution (i.e., endemic equilibrium).
From Figure 2, we can see the influence of transmission rate  and the average infectious period ⟨⟩.In particular, the influence of the diversity of the infectious periods is presented in Figure 2; the longer the infectious periods that the individuals have, the greater the basic reproduction number  0 is.When  is fixed,  0 is a monotone increasing with the value of ⟨⟩ and  0 is a monotone increasing function of  when ⟨⟩ is fixed.At the same time,  0 is linear function in terms of  and ⟨⟩.The basic reproduction number will increase with the increase of average infectious period and transmission rate.
We simulate the time series of total number of infected nodes on BA scale-free networks in Figure 3, which corresponds to  0 < 1 and  0 > 1, respectively.We can see that if  0 < 1, the disease will disappear quickly; otherwise when  0 > 1 the disease will persist in the system.So it is verified that  0 is the threshold for the dynamics of disease.

Conclusion and Discussion
In this paper, we establish an SIS epidemic spreading model with an arbitrary distribution of infectious period and take network structure into consideration.The disease-free equilibrium is globally asymptotically stable when  0 < 1.In the other case, there exists a unique endemic equilibrium such that it is globally attractive.Some numerical simulations are also performed to verify our theoretical results.
It is well known that, on a normal network, the basic reproduction number  0 depends on the degree distribution, transmission rate, and recovery rate.However, when the characteristics of different nodes have a larger difference, the  0 will depend on the distribution of these characteristics, such as the infectious period.From the equation of epidemic threshold, we have shown that the basic reproduction number  0 depends on the heterogeneity of the social networks and the diversity of the infectious periods of the individuals.And we obtain that the heterogeneity of the network and the long infectious period resulting in the infection deteriorate into endemic more easily.
By modifying the staged progression model, we propose the multistaged progression model which contains several different gamma distributions.The linear combination of gamma distributions with different parameters can describe an arbitrarily distributed distribution of the infectious period.We find that the number of stable infected individuals for different infectious periods is the same; however, the cumulative numbers of the infected individuals are different corresponding to the different infectious period distributions.And numerical simulations show that different infectious period distributions can lead to different transmission processes.Hence our model can characterize the diversity of the infectious period during the disease transmission on complex networks more realistic.

( 4 )
In order to analyze the model simply and efficiently, all transmission rates from infected individuals to susceptible individuals are .The duration of infected individuals  (,) in compartment (, )th is 1/.

Figure 1 :
Figure 1: The flowchart of disease spreading.

Figure 2 :
Figure 2:  0 as a function of  and ⟨⟩, which depends on the heterogeneity of the social networks and the diversity of the infectious periods of the individuals.