JAM Journal of Applied Mathematics 1687-0042 1110-757X Hindawi 10.1155/2019/8465747 8465747 Research Article Parameter Estimation and Sensitivity Analysis of Dysentery Diarrhea Epidemic Model http://orcid.org/0000-0001-8260-8885 Berhe Hailay Weldegiorgis 1 http://orcid.org/0000-0003-4550-9591 Makinde Oluwole Daniel 2 Theuri David Mwangi 3 Cao Jinde 1 Pan African University Institute of Basic Sciences Technology and Innovation Nairobi Kenya jkuat.ac.ke 2 Faculty of Military Science Stellenbosch University Cape Town South Africa sun.ac.za 3 Department of Mathematics Jomo Kenyatta University of Agriculture and Technology Nairobi Kenya jkuat.ac.ke 2019 1222019 2019 03 08 2018 23 12 2018 1222019 2019 Copyright © 2019 Hailay Weldegiorgis Berhe et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

In this paper, dysentery diarrhea deterministic compartmental model is proposed. The local and global stability of the disease-free equilibrium is obtained using the stability theory of differential equations. Numerical simulation of the system shows that the backward bifurcation of the endemic equilibrium exists for R0>1. The system is formulated as a standard nonlinear least squares problem to estimate the parameters. The estimated reproduction number, based on the dysentery diarrhea disease data for Ethiopia in 2017, is R0=1.1208. This suggests that elimination of the dysentery disease from Ethiopia is not practical. A graphical method is used to validate the model. Sensitivity analysis is carried out to determine the importance of model parameters in the disease dynamics. It is found out that the reproduction number is the most sensitive to the effective transmission rate of dysentery diarrhea (βh). It is also demonstrated that control of the effective transmission rate is essential to stop the spreading of the disease.

1. Introduction

Diarrheal disease is a common threat worldwide, particularly in developing countries. It is preventable and treatable. According to , diarrhea in its various forms is usually one of the five major causes of deaths in developing nations. It is the second leading cause of death in children under five years of age. Pneumonia and diarrhea are responsible for an estimated 40 percent of all child deaths each year. The authors  also reported that diarrhea causes 18% of all the deaths of children under the age of five. According to the study, more than 5000 children are dying every day as a result of diarrheal diseases. African and South-East Asian regions share 78% of all these deaths.

In most cases, diarrhea disease occurs in three forms and frequently occurs in children of under five years age [1, 3]. The first type of diarrhea is called dysentery diarrhea. This is diarrhea with the loose or watery stools with visible blood. Dysentery diarrhea is most often caused by shigella dysenteriae serotype 1 (bacillary dysentery) or Entamoeba histolytica (amoebic dysentery). The second form of diarrhea is acute watery diarrhea. This form includes cholera and is associated with significant fluid loss and rapid dehydration in an infected individual. It usually lasts for several hours or days. This can be caused by the pathogens V. cholerae or E. coli bacteria and rotavirus. The last form of diarrhea is persistent diarrhea. This is diarrhea with or without blood that lasts at least 14 days. It is common in undernourished children and those with other illnesses, such as AIDS.

Stochastic and deterministic mathematical models are widely employed by public health practitioners in order to predict and control disease outbreak as well as determining the cost effectiveness of the available strategies . In stochastic model, the disease transmission dynamics within a given population is based on the probabilistic analyses, while the deterministic models deals with the use of continuous functions to describe the time evolution of disease in a homogeneously mixed population . Several authors  have made use of stability theory of differential equations in order to analysis deterministic compartmental models for both local and global dynamics of diseases transmission in a within hosts and between hosts population such as HIV/AIDs, cholera, sheep brucellosis, influenza, typhoid, etc. In all these work, the threshold parameters for disease control and elimination were determined.

Epidemiologically speaking, it is important to have adequate knowledge of the physical characteristic of any given ailment or disease in order to provide accurate model that can be utilized to predict and control its outbreak effectively. The availability of relevant data for the disease will also enhance the model parameters estimation and the usefulness of the model with respect to the disease involved. Model parameters from epidemiological models can be easily estimated using methods involving ordinary least squares estimator, maximum likelihood estimator derivative approximation, moments estimator, Markov chain Monte Carlo (MCMC) strategy, derivative-free optimization algorithms, the Levenberg-Marquardt, and Trust-Region-Reflective . Biegler and Grossmann  employed optimization technique based on the seasonal data with the SIRS epidemic model as a constraint in order to estimate the parameters of a generalized incidence rate function. Li  also made use of optimization method with difference equations associated with a given system as a constraint in order to estimate the values of embedded parameters. The SIR model parameters were numerically estimated by Capaldi et al.  using least squares method. Other authors  have made use of various parameter estimation techniques base on the available data to determine the effects of various epidemiological factors on the disease transmission and possible control strategy. However, there is still a need to estimate the parameters involved in the various epidemiological models in order to assess the relative influence of each input variable on the output variable.

To the best of our knowledge problem involving estimation of parameters in dysentery diarrhea epidemic model with sensitivity analysis of the basic reproduction number has been scarcely investigated. The main objective of this paper is to tackle this problem by employing constrained optimization technique in order to estimate the values of parameters involved in a deterministic compartmental model for dysentery diarrhea epidemic as a constraint. Real data based on the weekly dysentery diarrhea epidemic that occurred in Ethiopia in 2017 were used to estimate parameters which are not accessible from the characteristic of the disease. In addition, we establish the local and global stability of the disease-free equilibrium and perform the bifurcation analysis. Sensitivity and uncertainty of the basic reproduction number of the system are analyzed. The study estimates the reproduction number of Ethiopia using the estimated parameters to predict the disease spread in the community. In the following Section 2, we present the model and its analysis. The numerical approximation and parameter estimation methods are presented in Section 3. In Section 4, numerical simulation of the system and sensitivity analysis with given parameter values is performed. Brief discussion and conclusions are presented in the last section.

2. The Model and Its Analysis

The SIRSB (Susceptible-Infected-Recovered-Susceptible-Pathogen population) model for dysentery diarrhea proposed by same authors  consists of the following system of nonlinear ordinary differential equations:(1)dSdt=Λ+αR-λh+λB+μS,dIdt=λh+λBS-μ+γ+dI,dRdt=γI-μ+αR,dBdt=ϵI-σB,S0=S00,E0=I00,R0=R00,B0=B00.The standard (frequency dependent) incidence in the human to human interaction and the logistic incidence in the environment to human interaction are respectively given by(2)λh=βhINand  λβ=βBBK+B.K is the shigella concentration that yields 50% chance of catching dysentery diarrhea . βB and βh represent rates of ingesting shigella from the contaminated environment and through human to human interaction, respectively. It is assumed that the incidence λB=βBB/K+B is a nonlinear function in B. This incidence represents a Hill function. If BK, λB grows linearly with B; if BK, λB approaches a constant state, βB, resulting in a saturation state. Therefore, this form is based on the assumption that the incidence has a saturation form. The pathogen population is diminished (at a rate of σ2-σ1=σ>0) [17, 35, 36].

The corresponding flow diagram and the description of the parameters for the model are given in Figure 1 and Table 1, respectively.

Description of parameters of the model equation (1).

Parameters Interpretation Units
Λ Recruitment rate of susceptible population Humans Time-1
μ Natural death rate of humans Time-1
γ Natural recovery rate of diarrhea Time-1
α Relapse rate of the recovered humans to the susceptible class T i m e - 1
d Disease induced death rate of dysentery diarrhea T i m e - 1
K Concentration of Shigella cells
βh Effective transmission rate of diarrhea due to human to human interaction T i m e - 1
βB Effective transmission rate of dysentery diarrhea due to environment to human interaction T i m e - 1
ϵ Pathogen shedding rate Cells Human-1Time-1
σ1 Shigella Pathogen growth rate T i m e - 1
σ2 Shigella Pathogen death rate T i m e - 1
σ Net death rate of Shigella Pathogen T i m e - 1

Schematic diagram for the flow of dysentery diarrhea infection.

2.1. Basic Reproduction Number and Existence of Equilibria

System equation (1) has one disease-free equilibrium E0=(Λ/μ,0,0,0). The basic reproduction number (R0) is utilized to quantify the transmission capability of a sickness. It is thought of as the quantity of secondary diseases delivered by a primary case of an infection in a population that is completely susceptible. It can in this manner be estimated by checking the quantity of secondary cases following the presentation of a disease into an absolutely susceptible population. R0 decides if a pathogen can hold on in such a population, and it is important for evaluating control alternatives. At the point when R0 is under 1, on average each infected individual infects less than one other individual, and the pathogen will cease to exist in the population. Conversely, when R0 surpasses 1 there is an exponential ascent in the quantity of cases after some time and disease epidemic results . Using the next generation approach of , the reproduction number is(3)R0=βhμ+γ+d+ΛβBϵμμ+γ+dKσ.The endemic equilibrium E=(S,I,R,B) of the system is the solution of the algebraic equation(4)Λ-λh+λB+μS+αR=0,λh+λBS-μ+d+γI=0,γI-μ+αR=0,ϵI-σB=0.That is,(5)B=ϵIσ,R=γIα+μ,S=Λα+μ-μμ+γ+d+αμ+dIμα+μ  andI is the solution of the following quadratic equation:(6)AI2+BI+C=0,

where(7)A=ϵμβh-βBdμμ+d+γ+αμ+d-dμ+d+γ,B=Λϵα+μβBd-μβh+μμ+d+γ+αμ+dKσ+ϵΛβB-μ+d+γKdσ+ϵΛ,C=μΛKσα+μμ+d+γ1-R0.

System equation (1) always has the disease-free equilibrium.

If R0>1 and A<0, B>0, then system equation (1) may has two endemic equilibria E1 and E2.

If R0>1, and A<0, B<0, then system equation (1) may has a unique endemic equilibrium, E.

If R0>1 and A>0, B<0, then system equation (1) may has a unique endemic equilibrium, E.

If R0>1 and A>0, B>0, then system equation (1) has no endemic equilibrium.

If 0R0<1, there is no endemic equilibrium.

2.2. Stability of the Disease-Free Equilibrium

This section is devoted to analytic conditions for the stability of the disease-free equilibrium.

2.2.1. Local Stability of the Disease-Free Equilibrium Theorem 1.

The disease-free equilibrium E0 is locally asymptotically stable if R01 and is unstable if R0>1.

Proof.

Local stability is analyzed by the Jacobian matrix of (1) at E0. That is, (8)JE0=-μ-βhα-βBΛKμ0βh-μ+d+γ0βBΛKμ0γ-α+μ00ϵ0-σ.The eigenvalues of this matrix are(9)λ1=-μ,λ2=-μ+αand  the  solutions  of  λ2+a1λ+a2=0where  a1=μ+d+γ+σ-βhand  a2=μ+d+γσ1-R0.If R0<1, a1<0 and a1a2>0. Thus, according to the Hurwitz criterion, the quadratic equation, λ2+a1λ+a2=0, has negative real eigenvalues. Hence, E0 is locally asymptotically stable. If R0=1, one eigenvalue of the quadratic equation is 0 and it is simple. Hence, based on Hartman-Grobman Theorem , the disease-free equilibrium E0 is a nonhyperbolic equilibrium. The loss of hyperbolicity of the equilibrium implies that the local behavior of the system cannot be described by the linearized system. And instead center manifold theory has to be used. The bifurcation analysis Theorem 3 below proves that E0 is the only equilibrium at R0=1 and hence E0 is locally asymptotically stable. If R0>1, then E0 is unstable.

2.2.2. Global Stability of the Disease-Free Equilibrium

In this section, we use the method used in  to investigate the global asymptotic stability of the disease-free equilibrium. To do so, we write system equation (1) as(10)dXdt=FX,I,dIdt=GX,I,GX,0=0,where XRm denotes the number of noninfectious individuals and IRn denotes the number of infected individuals. Let U0=(X,0) denote the disease-free equilibrium of this system. Furthermore, suppose that the following assumptions are true:

(H1) For dX/dt=F(X,0), X is globally asymptotically stable;

(H2) G(X,I)=AI-G^(X,I), G^(X,I)0 for (X,I)A,

where A=DIG(X,0) is an M-matrix.

Theorem 2.

The fixed point U0=(X,0) is a globally asymptotic stable equilibrium of model equation (1) provided that R01 and that assumptions (H1) and (H2) hold.

Proof.

To prove this, we write (1) in the form of (10). That is, X=S, I=(I,B). Furthermore, (11)FX,0=Λ-μS0,A=βh-μ+γ+dβBΛKμϵ-σand(12)G^X,I=G1^X,IG2^X,I=1-SNβhI+ΛKμ-SK+BβBB0.Since 1-S/N>0 and Λ/Kμ-S/K+B>0, then G^(X,I)0. It is also clear that X=(Λ/μ,0) is a global stable equilibrium of the system dX/dt=F(X,0). Hence, the above theorem U0=(X,0) is globally stable.

2.3. Bifurcation Analysis Theorem 3.

If R0<1, the system exhibits a forward bifurcation at R0=1.

Proof.

To investigate the type of bifurcation, let(13)βBμKσϵΛμ+d+γ-βhbe a bifurcation parameter. The Jacobin matrix at E0 with βB=βB is (14)JE0=-μ-βhα-βBΛKμ0βh-μ+d+γ0βBΛKμ0γ-α+μ00ϵ0-σ.The eigenvalues of the characteristics polynomial at R0=1 are λ1=-μ, λ2=-(α+μ), λ3=βhΛ/μ-(μ+d+γ+σ), and λ4=0. It can observe that the three eigenvalues are real and negative and one is 0 and simple. Now we denote by w=(w1,w2,w3,w4)T the right eigenvector corresponding to λ4=0. Direct calculation gives, (15)w=-αμ+d+μμ+d+γμα+μw2,w2,ϵσw2,γμ+αw2T.Moreover, the left eigenvector v=(v1,v2,v3,v4)T that satisfy v.w=1 is given by (16)-μv1=0,-βhv1+βh-μ+d+γv2+γv3+ϵv4=0,αv1-μ+αv3=0,-σϵμ+d+γ-βhv1+σϵμ+d+γ-βhv2-σv4=0.Then, the left eigenvector v turns out to be (17)v=0,v2,0,μ+d+γ-βhϵv2T.The equation v.w=1 results in v2w2+v4w4=1. Assume w2=ϵ(μ+α). Then (18)v=0,1ϵμ+α+γμ+d+γ-βh),0,μ+d+γ-βhϵϵμ+α+γμ+d+γ-βh)T.w=-αμ+d+μμ+d+γμϵ,ϵμ+α,ϵ2σμ+α,γϵT.Based on theoretical results in , we have to compute a and b where(19)a=k,i,j=13vkwiwj2fkxixjE0,βB,b=k,i=13vkwi2fkxiβE0,βB.Set x1=S, x2=I, x3=R, x4=B. Evaluating the partial derivatives at (E0,βB), we obtain (20)2f2x1x2E0,βB=2f2x2x1E0,βB=βhS0,2f2x1x4E0,βB=2f2x4x1E0,βB=-βBK2f2x42E0,βB=-2βBK2.Furthermore, (21)2f2x3βBE0,βB=ΛμKand all the other second order partial derivatives are zero. It follows that (22)a=2v2w1w22f2x1x2E0,βB+w1w42f2x1x4E0,βB+w4222f2x42E0,βBand(23)b=v2w42f2x4βBE0,βB.Substituting the eigenvectors and the partial derivatives into a and b, we get(24)a=-2ϵμ+αΛϵμ+α+γμ+d+γ-βh)a0where  ao=αμ+d+μμ+d+γβh+αμ+d+μ(μ+d+γ+γα+μσϵγα+μμ+d+γ-βhand(25)b=1ϵμ+α+γμ+d+γ-βhγϵΛμK.

If R0<1, then (μ+d+γ)-βh>0, a<0 and b>0. This scenario indicates that the system exhibits transcritical bifurcation at βB=βB. Its biological meaning is that as long as R0<1 the dysentery epidemic can be eliminated from the population. If parameters change and result in R0>1, a small endemic state may occur.

3. Numerical Methods 3.1. Parameter Estimation

In this manuscript we solve a dynamics parameter estimation problem. It aims to solve for the unknown parameters of a dynamic model supplied as a system of ordinary differential equations. The problem is setup as a standard nonlinear least squares problem; however the nonlinear function to be fitted involves solving ordinary differential equations using a numerical integration scheme. Least squares are a special case of maximum likelihood in which it is assumed that the data are drawn from a normal distribution with mean given by the solution to the ODE. System equation (1) is formulated as a standard dynamic model in the following form:(26)y=ft,y,ϑ,yt0=y0,where y is the vector of dependent variables and ϑ is the vector of unknown parameters. The error is sum-of-squares error and is represented by(27)ψϑ=i=1yi-yi¯2in which yi¯ is the real data and yi=y(ti,ϑ) is the solution to (26) at time ti for a given ϑ. The aim is to minimize the objective function(28)minϑψϑsubjecttoEq.  26to obtain our parameter estimates. The parameter estimation algorithm based on  with some modification is summarized below.

Algorithm

Result: Optimal estimated parameter values ω.

Guess initial parameter values ω0. Set ω=ω0.

Using MATLAB ode45 routine, solve (26) using ω to find the solution yi.

Evaluate error using (27).

Minimize (28) using an optimization algorithm lsqnonlin along with Trust-Region-Reflective to find bounded estimated parameters ω^. Update ω=ω^.

Check convergence criteria. If not converged, go to (2).

On convergence, set ω=ω^ to current parameter values.

4. Numerical Simulations and Discussion

In this section, we first present the existence and nonexistence of the endemic equilibria. Secondly, fitting the data of dysentery diarrhea cases in Ethiopia is performed to estimate the parameters. Next, the sensitivity of the R0 to the system parameters is analyzed. Finally, the uncertainty analysis is presented using the estimated parameters.

4.1. The Disease-Free and Endemic Equilibria

It is proved in Theorem 2 that the disease does not exist in the community so long as the reproduction number is less than unity and the disease is endemic in the community if the reproduction number is greater than unity. It is depicted in Figure 2(a). The disease-free equilibrium is globally stable if R01. However, there exists a possibility of multiple equilibria for R0>1 (Figure 2(b)).

(a) Forward bifurcation graph for Λ=0.5, βB=0.4, K=1, μ=0.2, ϵ=0.05, d=0.1, γ=0.1, σ=3, α=1, and βh varies βh=0.001:0.01:15. (b) Multiple equilibria for Λ=2, βB=5, K=1, μ=0.5, ϵ=0.05, d=0.8, γ=0.1, σ=3, α=1, and βh varies βh=0:0.001:50.

Forward bifurcation

Multiple equilibria for R0>1

4.2. Dysentery Diarrhea Data

Dysentery diarrhea is one of the most dangerous types of diarrhea which is endemic in Ethiopia. It is reported weekly under Integrated Disease Surveillance and Response System (IDSR) in the Ministry of Health. Ethiopian Weekly Epidemiological Bulletin  reports National Surveillance Data Summary of diseases in Ethiopia. For instance, dysentery surveillance data of week 10 of 2018 is 5,345. There is an increment of 614 as compared to week 9, 2018. However, the number of cases reported during this week is lower than the same weeks of the previous two years. The data of dysentery diarrhea reported cases in 2017 used in this simulation are presented in (Table 2). Time series graphs of data given in (Table 2) are shown in Figure 3. As shown in Figure 3, the infected cases have maximum amplitude in the sixteenth week.

Weekly dysentery diarrhea reports in Ethiopia, 2017.

Weeks Infected Weeks Infected Weeks Infected Weeks Infected Weeks Infected
1 4542 12 7000 23 7375 33 7042 43 5583
2 4750 13 7042 24 7167 34 6917 44 5458
3 4792 14 6458 25 7292 35 6208 45 5458
4 5417 15 7083 26 7000 36 5958 46 5458
5 5250 16 7625 27 6583 37 6500 47 5250
6 5125 17 6875 28 6750 38 5958 48 4750
7 5833 18 6375 29 6833 39 5583 49 4750
8 6000 19 6625 30 6792 40 5667 50 4750
9 5542 20 6875 31 6708 41 5667 51 4583
10 5917 21 7000 32 6833 42 5542 52 4500
11 6583 22 7042

Weekly time series of dysentery diarrhea data in Ethiopia 2017.

4.3. Estimation of Parameters

In this model we have ten parameter values to be estimated. Among these parameters, the value of μ, natural mortality rate is obtained from the local demography of Ethiopia. The parameter γ, the duration of infectiousness, is also found from dysentery diarrhea characteristics . The initial infectious individuals are taken as the number of infected ones in the first week of 2017. Furthermore, the shigella concentration that yields 25-50% chance of catching dysentery diarrhea is 200 . Thus, the initial parameter values for estimation is ω0=(Λ,βh,βB,K,μ,γ,d,α,ϵ,σ)=(2500,0.0158,0.052,200,0.000457,0.1106,0.0103,0.018, 0.01,0.3319) and the initial population is (S0,I0,R0,B0)=(416253,4542,200,2000). Using the algorithm in Section 3.1 the estimated parameter values are given in Table 3.

Estimated parameter values.

Parameter Value Bounds Source
Λ 462 [ 0,84388.8700 ] Estimated
β h 0.69864 [ 0.2175,1.1798 ] Estimated
β B 0.04465 [ 0,2.4170 ] Estimated
K 200 
μ 0.000457 
γ 0.1106 
d 0.55602 [ 0.0581,1.0540 ] Estimated
α 0.0211 [ 0,1.0588 ] Estimated
ϵ 0.01 Assumed
σ 46.2086 [ 0,2520.3600 ] Estimated
S 0 216253 [ 91252,341254 ] Estimated
I 0 3152 [ 0,58153 ] Estimated
R 0 165 [ 0,25288 ] Estimated
B 0 1356 [ 0,4520 ] Estimated

The reproduction number based on the Ethiopian data is therefore R0=R2+R1=1.0476+0.0732=1.1208. R2 is the reproduction number due to the human to human interaction and R1 is the reproduction number due to the environment to human interaction.

4.4. Sensitivity Analysis

The sensitivity analysis reveals how imperative every parameter is to illness transmission. It is regularly used to decide the robustness of model expectations to parameter values since there are errors in data collection and assumed parameter values . The perturbation of fixed point estimation of model parameters and the uncertainty in the model parameter estimation are the two most commonly used techniques for sensitivity analysis. The sensitivity of a variable with respect to system parameters is usually measured by sensitivity index. Sensitivity indices enable us to quantify the relative change in a variable when a parameter changes. The normalized forward sensitivity index of a variable regarding a parameter is the proportion of the relative change in the variable to the relative change in the parameter. When the variable is a differentiable function of the parameter, the sensitivity index might be on the other hand defined utilizing partial derivatives.

Definition 4 (see [<xref ref-type="bibr" rid="B42">42</xref>, <xref ref-type="bibr" rid="B43">43</xref>]).

The normalized forward sensitivity index of R0 that depends differentiably on a parameter ρ is defined by (29)φρR0=R0ρρR0.

For instance, φρR0=1 implies an increase (decrease) of ρ by y% increases (decreases) R0 by the same percentage. On the other hand, φρR0=-1 indicates an increase (decrease) of ρ by y% decreases (increases) R0 by y%. In this case, ρ is called a highly sensitive parameter.

Proposition 5.

Given(30)R0=βhμ+γ+d+ΛβBϵμμ+γ+dKσ.Let(31)R2=βhμ+γ+dand  R1=ΛβBϵμμ+γ+dKσ.The normalized forward sensitivity index of R0 with respect to the given parameters is given by (32)φβhR0=R2R0,φdR0=-dμ+d+γ,φγR0=-γμ+d+γ,φμR0=-μ2μ+d+γ+R1R0,φβBR0=φϵR0=φΛR0=R1R0,φKR0=φσR0=-R1R0.

Proof.

This follows immediately from Definition 4.

The sensitivity indices of φΛR0, φβhR0, φβBR0, φϵR0 are positive and the remaining are negative. Since all indices are the functions of other parameters the sensitivity indices will change with the change in other parameter values. The parameter values in Table 3 are used to calculate the sensitivity indices. The normalized sensitivity of the basic reproduction number with respect to the estimated parameters is given in Table 4. For example, φβhR0=0.9901115761; the physical meaning is that increasing (or decreasing) βh by 10% increases (or decreases) R0 by 9.9%. A highly sensitive parameter should be carefully estimated, because a small variation in that parameter will lead to large quantitative changes. As can be seen from Table 4, the basic reproduction number is most sensitive to the effective transmission rate of dysentery diarrhea (βh). The implication of this is that an increase in the transmission rate increases the spread of the disease in the community.

Estimated parameters.

Parameter Sensitivity indices of R0
Λ φΛR0=0.009888424079
βh φ β h R 0 = 0.9901115761
βB φ β B R 0 = 0.009888424076
K φ K R 0 = - 0.009888424073
μ φ μ R 0 = - 0.01466434116
γ φ γ R 0 = - 0.1651192337
d φ d R 0 = - 0.8301048495
ϵ φ ϵ R 0 = 0.009888424080
σ φ σ R 0 = - 0.009888424073
4.5. Model Validation

Once a model is developed, we use real data to validate it. There are many statistical techniques applied for model validation. The two most frequently used methods are graphical and numerical techniques. Graphical methods demonstrate a comprehensive range of complex features of the relationship between the model and the real data. A numerical technique for model validation, on the other hand, tends to be narrowly targeted on a specific side of the connection between the model and the observed data.

4.5.1. Graphical Method

The residuals from a fitted model are the differences between the real data and the solution to the model. If the model fits the real data appropriately, the residuals approximate the random errors that make the association with the real data and the solution to the system . As a result, if the residuals seem to behave randomly, it is sound to say that the model fits the real data. On the contrary, if nonrandom arrangement is visible in the residuals, the model fails to depict the true representation of the real data. Keeping in mind the goal to check the validity of the model to the given information, Figure 4(a) has been drawn. Figure 4(a) demonstrates the results of fitting the SIRSB model in (1) to the dysentery data in Table 2. For this figure, the continuous line represents the model output and the points represent the real data points. There are a few data points that fit poorly with the model outcome. However, overall the model solution and data agree well.

Fitting of the model equation (1) to the reported data in Table 2 and residual analysis using the estimated parameters.

Fitting

Residuals

Residuals plots for the model and the real data in Table 2 are given in Figure 4(b). As given in Figure 4(b), the residuals seem to be random and the standard errors are generally small for the model. In this manner the estimates acquired here are reasonable.

The residual analysis is determined by computing autocorrelations for the residuals at varying time Lags. Residual analysis includes two tests: the whiteness test and the independence test. The whiteness test is about autocorrelation function of the residuals for each output. Based on this test, the residual autocorrelation function of a good model is inside the confidence interval of the corresponding estimates, showing that the residuals are uncorrelated. On the other hand, the independence test checks the cross-correlation between the input and the residuals for each input output pair. The independence test makes evident that a good model has the residuals uncorrelated with past inputs. In this case, correlation shows that the model does not describe how part of the output relates to the corresponding input.

The autocorrelation function plot for the residuals shows that most of the residuals are not statistically significant (Figure 5) which directly follows that the residuals are not highly correlated. Except for one major downturn, the residuals between week one and the last week do not show any particular form. They fluctuate randomly around zero. The independence of the residuals obtained suggests that the parameter estimates obtained are reliable. It is easy to see from the plot that almost all of the autocorrelations except the fifth Lag fall within the 99% confidence limits. One Lag outside the 99% confidence limits does not necessarily show nonrandomness. Furthermore, adjacent observations do not correlate.

Correlation of residuals for the infected individuals Λ=462, μ=0.003199, βh=0.69864, βB=0.04465, K=200, α=0.0211, ϵ=0.01, γ=0.1106, σ=46.2086, and K=0.1.

4.5.2. Uncertainties Analysis

In the process of parameter estimation, it is reasonable to assume that the parameters are dependent. Particularly, the parameters could be thought to vary together or covary. Covariance is the measure of how the parameter yi is associated with yj. In the analysis of model parameters a covariance matrix approximates the uncertainties in the model parameters and the possible disorders in the outputs. The diagonal elements of the matrix contain the variances of the parameters and the off-diagonal elements contain the covariances between all potential pairs of parameter. This matrix Σ is positive semidefinite and symmetric. The interrelations between the estimated elements in this are based on the signs. If the covariance between any two estimated parameters is positive, then the parameters values tend to vary in a positive way. If the covariance between any two estimated parameters is 0, then the two parameters are unrelated. On the other hand, if the covariance between any two parameters is negative, then the parameters values tend to move in opposite directions. The covariance matrix Σ for the model is represented below. (33)Σ=βhβBdΛασβh0.230.160.2434654-0.42160βB0.165.630.1552.170.0225869.23d0.240.150.2536864.41-0.45153.64Λ34654521.6536864.407043651412.80-87035.24-762050.23α-0.420.022094-0.45-87035.241.0839.46σ1605869.23153.64-762050.2239.466121430.51.The variance in descending order are recruitment rate of susceptible population, Λ (7043651412.8), the net death rate of shigella pathogen, σ (6121430.51), effective transmission rates of dysentery diarrhea due to the environment to human interaction, βB (5.63), relapse rate of the recovered ones to susceptible, α (1.08), disease induced death rate of dysentery diarrhea, d (0.23), and effective transmission rates of diarrhea due to the human to human interaction, βh (0.23).

The covariance between transmission coefficient (βh) and transmission coefficient (βB), recruitment rate of susceptible population (Λ), disease induced death rate of dysentery diarrhea (d), and net death rate of shigella pathogen (σ) is positive. The covariance between transmission coefficient, βB and recruitment rate of susceptible population (Λ), disease induced death rate of dysentery diarrhea (d), net death rate of shigella pathogen (σ), and relapse rate of the recovered ones to susceptible (α) is positive. The covariance between recruitment rate of susceptible population (Λ) and disease induced death rate of dysentery diarrhea (d) is positive. The possible implication of this is that if one parameter is increased, the other parameter must also be increased to achieve a similar fit to the data. On the other hand, the covariance between the relapse rate of the recovered ones to susceptible class (α) and transmission coefficient (βh), disease induced death rate of dysentery diarrhea (d), and recruitment rate of susceptible population (Λ) is negative. The covariance between transmission coefficient (βh) and the relapse rate of the recovered ones to susceptible class (α) is negative. The covariance between recruitment rate of susceptible population (Λ) and the relapse rate of the recovered ones to susceptible class (α) and net death rate of shigella pathogen (σ) is negative. The physical implication of this is that if one parameter is increased, the other parameter must be decreased to achieve a similar fit to the data. Table 5 shows the summary of covariance of the model parameters.

Covariance of the parameters.

Covariance Direction Covariance Direction Covariance Direction
Cβh,βB +ve C β B , d +ve C α , d -ve
Cβh,d +ve C β B , Λ +ve C α , Λ -ve
Cβh,Λ +ve C β B , α +ve C α , σ +ve
Cβh,α -ve C β B , σ +ve C β h , σ +ve
C σ , d +ve C σ , Λ -ve
5. Discussion and Conclusions

In this paper, we have proposed a deterministic compartmental dysentery diarrhea model. We have proved the global stability of the disease-free equilibrium. It is established in Theorem 2 that the basic reproduction number R0 is a sharp threshold parameter and it completely determines the global stability of the disease-free equilibrium for R0<1. The possible implication of this is that to eliminate the disease from the community, policymakers have to work on reducing the reproduction number to below unity.

The authors  have reported that the contribution of each transmission pathway in a disease outbreak is separable to focus on the control efforts of the infectious humans or the contaminated environment. Accordingly, R0 could be written as(34)R0=βhμ+d+γ+ΛβBϵμμ+d+γKσ21-σ1/σ2.Using Taylor’s series expansion, we have(35)1-σ1σ2-1=1+σ1σ2+σ1σ22+higher  order  terms.Thus, (36)R0=R01+R02+R03+higher  order  terms.where R01=βh/μ+d+γ, R02=ϵΛ/μ1/μ+d+γβB/K1/σ2, and R03=ϵΛ/μ1/μ+d+γβB/K1/σ2σ1/σ2.

R 01 can be expressed as a primary case in the human population makes an infectious contact with humans at a rate of βh over the mean infectious period of 1/μ+d+γ. R02 can be interpreted as a primary case in the human population makes an infectious contact with the environment at a rate of ϵΛ/μ over the mean infectious period of 1/μ+d+γ and a primary case in the environment makes an infectious contact with humans at a rate of βB/K over the mean infectious period of 1/σ2. R03 is interpreted as a primary case in the human population makes an infectious contact with the environment at a rate of ϵΛ/μ over the mean infectious period of 1/μ+d+γ and a primary case in the environment makes an infectious contact with humans at a rate of βB/K over the mean infectious period of 1/σ2, a fraction σ1/σ2 of which survive and become infectious.

Therefore, R0=R01+R02+R03 represents the total contribution to the infectious class and the pathogen population class made by the hosts and pathogens of original case. The higher order terms represent the contribution through direct human to human and environment to human in the higher generation.

As an application, we have used our system to simulate the reported dysentery diarrhea cases in Ethiopia in 2017 (Table 2) and obtained reasonable matches. To solve the optimization problem, the Runge-Kutta method (ode45) has been used as a solver of a system of nonlinear differential equations along with Trust-Region-Reflective (to find optimum bounded parameter values) and lsqnonlin methods for the sum of squared residuals. The results indicate that simulations of our system can provide a match to the infectious cases in 2017. More importantly, we estimate that the basic reproduction number for dysentery transmission in Ethiopia is R0=1.1208. The implication of this is that dysentery diarrhea is endemic in country. The estimated reproduction number is near to one. If control and treatment measures are implemented, with this value of R0, the disease will be eliminated in short time.

The values of parameters describing the system have been estimated by fitting the integrals of the system to the field data on dysentery diarrhea epidemic. Estimation of parameters, in this case, is a challenging task because of missing a large part of the infectious process. One difficulty is that depending on the initially specified parameter values a local minimum may occur and the minimal value of the sum of squared residuals may be different. In such case, the parameter estimation should be repeated with different initial parameter guesses to achieve a better estimate of the global minimum (rather than local) of the sum of squared residuals.

Data Availability

The data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

United Nations International Children’s Emergency Fund Diarrhea: Why Children Are Still Dying and What Can Be Done 2009 Farthing M. Salam M. A. Lindberg G. Dite P. Khalif I. Salazar-Lindo E. Ramakrishna B. S. Goh K.-L. Thomson A. Khan A. G. Krabshuis J. Lemair A. Acute diarrhea in adults and children: a global perspective Journal of Clinical Gastroenterology 2013 47 1 12 20 10.1097/mcg.0b013e31826df662 2-s2.0-84871572078 Bopp C. A. Ries A. A. Wells J. G. Laboratory Methods for the Diagnosis of Epidemic Dysentery and Cholera 1999 Hugo A. Makinde O. D. Kumar S. Chibwana F. F. Optimal control and cost effectiveness analysis for Newcastle disease eco-epidemiological model in Tanzania Journal of Biological Dynamics 2017 11 1 190 209 10.1080/17513758.2016.1258093 MR3733432 2-s2.0-85034587078 Olaniyi S. Dynamics of Zika virus model with nonlinear incidence and optimal control strategies Applied Mathematics & Information Sciences 2018 12 5 969 982 2-s2.0-85053391273 10.18576/amis/120510 Anderson R. M. May R. M. Age-related changes in the rate of disease transmission: Implications for the design of vaccination programmes Epidemiology & Infection 1985 94 3 365 436 2-s2.0-0022262704 10.1017/S002217240006160X 4008922 Roberts M. G. Tobias M. I. Predicting and preventing measles epidemics in New Zealand: Application of a mathematical model Epidemiology and Infection 2000 124 2 279 287 2-s2.0-0034089305 10.1017/S0950268899003556 Li L. Wang C.-H. Wang S.-F. Li M.-T. Yakob L. Cazelles B. Jin Z. Zhang W.-Y. Hemorrhagic fever with renal syndrome in China: mechanisms on two distinct annual peaks and control measures International Journal of Biomathematics 2018 11 2 1850030 10.1142/S1793524518500304 MR3767049 Anderson R. M. May R. M. Infectious Diseases of Humans: Dynamics and Contro 1992 Oxford University Press Choisy M. Gugan J. F. Rohani P. Mathematical Modeling of Infectious Diseases Dynamics 2007 Athens, GA, USA Institute of Ecology, University of Georgia Brauer F. Castillo-Chavez C. Mathematical Models in Population Biology and Epidemiology 2012 1st Springer Texts in Applied Mathematics Lu G. Lu Z. Geometric approach to global asymptotic stability for the {SEIRS} models in epidemiology Nonlinear Analysis: Real World Applications 2017 36 20 43 10.1016/j.nonrwa.2016.12.005 MR3621228 Zbl1379.34043 Eisenberg M. C. Robertson S. L. Tien J. H. Identifiability and estimation of multiple transmission pathways in cholera and waterborne disease Journal of Theoretical Biology 2013 324 84 102 10.1016/j.jtbi.2012.12.021 MR3041644 Zbl1314.92156 2-s2.0-84875251690 23333764 Liao S. Yang W. On the dynamics of a vaccination model with multiple transmission ways International Journal of Applied Mathematics and Computer Science 2013 23 4 761 772 10.2478/amcs-2013-0057 MR3185182 Zbl1285.93043 2-s2.0-84893754714 Tien J. H. Earn D. J. Multiple transmission pathways and disease dynamics in a waterborne pathogen model Bulletin of Mathematical Biology 2010 72 6 1506 1533 10.1007/s11538-010-9507-6 MR2671584 Zbl1198.92030 2-s2.0-77955172112 Sun G.-Q. Zhang Z.-K. Global stability for a sheep brucellosis model with immigration Applied Mathematics and Computation 2014 246 336 345 10.1016/j.amc.2014.08.028 MR3265873 2-s2.0-84907214486 Zbl1338.92141 Bani-Yaghoub M. Gautam R. Shuai Z. van den Driessche P. Ivanek R. Reproduction numbers for infections with free-living pathogens growing in the environment Journal of Biological Dynamics 2012 6 2 923 940 10.1080/17513758.2012.693206 2-s2.0-84868316339 Misra A. K. Gupta A. Venturino E. Cholera dynamics with bacteriophage infection: a mathematical study Chaos, Solitons & Fractals 2016 91 610 621 10.1016/j.chaos.2016.08.008 MR3551749 Li L. Monthly periodic outbreak of hemorrhagic fever with renal syndrome in China Journal of Biological Systems 2016 24 4 519 533 10.1142/S0218339016500261 MR3591570 Zbl1360.92101 2-s2.0-85003955322 Woldegerima W. A. Teboh-Ewungkem M. I. Ngwa G. A. The impact of recruitment on the dynamics of an immune-suppressed within-human–host model of the plasmodium falciparum parasite Bulletin of Mathematical Biology 2018 1 56 2-s2.0-85047327096 Mbabazi F. K. Mugisha J. Kimathi M. Modeling the within-host co-infection of influenza A virus and pneumococcus Applied Mathematics and Computation 2018 339 488 506 10.1016/j.amc.2018.07.031 Schittkowski K. Numerical Data Fitting in Dynamical Systems 2002 77 Kluwer Academic Publishers Applied Optimization 10.1007/978-1-4419-5762-7 MR1957699 Kristensen R. M. Parameter Estimation in Nonlinear Dynamical System [MSc thesis] 2004 Department of Chemical Engineering, Technical University of Denmark Lei M. U. Parameter estimation methods for biological systems, [MSc thesis] 2010 Saskatoon, Canada Department of Chemical Engineering, University of Saskatchewan Samsuzzoha M. Singh M. Lucy D. Parameter estimation of influenza epidemic model Applied Mathematics and Computation 2013 220 616 629 10.1016/j.amc.2013.07.040 MR3091886 2-s2.0-84881269353 Capistran M. A. Moreles M. A. Lara B. Parameter estimation of some epidemic models. The case of recurrent epidemics caused by respiratory syncytial virus Bulletin of Mathematical Biology 2009 71 8 1890 1901 10.1007/s11538-009-9429-3 MR2551694 Biegler L. T. Grossmann I. E. Retrospective on optimization Computers & Chemical Engineering 2004 28 8 1169 1192 2-s2.0-2342632555 10.1016/j.compchemeng.2003.11.003 Li Z. Parameter estimation of ordinary differential equations IMA Journal of Numerical Analysis (IMAJNA) 2005 25 2 264 285 10.1093/imanum/drh016 Zbl1070.65061 Capaldi A. Behrend S. Berman B. Smith J. Wright J. Lloyd A. L. Parameter estimation and uncertainty quantification for an epidemic model Mathematical Biosciences and Engineering 2012 9 3 553 576 10.3934/mbe.2012.9.553 MR2957535 Zbl1260.62085 Yorke J. A. London W. P. Recurrent outbreaks of measles, chickenpox and mumps American Journal of Epidemiology 1973 98 6 469 482 10.1093/oxfordjournals.aje.a121576 2-s2.0-0015711463 Kalivianakis M. Mous S. L. J. Grasman J. Reconstruction of the seasonally varying contact rate for measles Mathematical Biosciences 1994 124 2 225 234 2-s2.0-0028606413 10.1016/0025-5564(94)90044-2 7833596 Zbl0818.92021 Longini I. M. Koopman J. S. Monto A. S. Fox J. P. Estimating household and community transmission parameters for influenza American Journal of Epidemiology 1982 115 5 736 751 2-s2.0-0019948714 10.1093/oxfordjournals.aje.a113356 7081204 Samsuzzoha M. D. Singh M. Lucy D. Uncertainty and sensitivity analysis of the basic reproduction number of a vaccinated epidemic model of influenza Applied Mathematical Modelling 2013 37 3 903 915 10.1016/j.apm.2012.03.029 MR3002196 2-s2.0-84870239692 Zbl1352.92168 Ojo O. B. Lougue S. Woldegerima W. A. Bayesian generalized linear mixed modeling of Tuberculosis using informative priors PLoS ONE 2017 12 3 e0172580 2-s2.0-85014305363 Berhe H. W. Makinde O. D. Theuri D. M. Optimal control and cost-effectiveness analysis for dysentery epidemic model Applied Mathematics and Information Sciences 2018 12 1183 1195 Berhe H. W. Makinde O. D. Theuri D. M. Co-dynamics of measles and dysentery diarrhea diseases with optimal control and cost-effectiveness analysis Applied Mathematics and Computation 2019 347 903 921 10.1016/j.amc.2018.11.049 MR3884704 Cabral J. P. S. Water microbiology. Bacterial pathogens and water International Journal of Environmental Research and Public Health 2010 7 10 3657 3703 2-s2.0-78649669767 10.3390/ijerph7103657 Melese Z. T. Mwalili S. M. Orwa G. O. Threshold dynamics of the transmission of antibiotic-resistant infections Biosystems 2018 171 80 92 10.1016/j.biosystems.2018.06.004 2-s2.0-85050665057 van den Driessche P. Watmough J. Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission Mathematical Biosciences 2002 180 1-2 29 48 10.1016/S0025-5564(02)00108-6 MR1950747 Castillo-Chavez C. Song B. Dynamical models of tuberculosis and their applications Mathematical Biosciences and Engineering 2004 1 2 361 404 10.3934/mbe.2004.1.361 MR2130673 Assefa Z. Bancha M. Ethiopian public health institute center for public health emergency management Ethiopian Weekly Epidemiological Bulletin 2018 4 10 6 7 Chitnis N. Hyman J. M. Cushing J. M. Determining important parameters in the spread of malaria through the sensitivity analysis of a mathematical model Bulletin of Mathematical Biology 2008 70 5 1272 1296 10.1007/s11538-008-9299-0 MR2421498 Zbl1142.92025 2-s2.0-47349102730 Silva C. J. Torres D. F. M. Optimal control for a tuberculosis model with reinfection and post-exposure interventions Mathematical Biosciences 2013 244 2 154 164 10.1016/j.mbs.2013.05.005 MR3101449 2-s2.0-84881227998 Guthrie W. NIST/SEMATECH e-Handbook of Statistical Methods 2000 National Institute of Standards and Technology