Complexity

Volume 2018, Article ID 3060368, 9 pages

https://doi.org/10.1155/2018/3060368

## Modeling Chickenpox Dynamics with a Discrete Time Bayesian Stochastic Compartmental Model

Correspondence should be addressed to F. J. Santonja; se.vu@ajnotnas.ocsicnarf

Received 1 September 2017; Revised 18 December 2017; Accepted 5 February 2018; Published 20 March 2018

Academic Editor: Dimitri Volchenkov

Copyright © 2018 A. Corberán-Vallet 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.

#### Abstract

We present a Bayesian stochastic susceptible-exposed-infectious-recovered model in discrete time to understand chickenpox transmission in the Valencian Community, Spain. During the last decades, different strategies have been introduced in the routine immunization program in order to reduce the impact of this disease, which remains a public health’s great concern. Under this scenario, a model capable of explaining closely the dynamics of chickenpox under the different vaccination strategies is of utter importance to assess their effectiveness. The proposed model takes into account both heterogeneous mixing of individuals in the population and the inherent stochasticity in the transmission of the disease. As shown in a comparative study, these assumptions are fundamental to describe properly the evolution of the disease. The Bayesian analysis of the model allows us to calculate the posterior distribution of the model parameters and the posterior predictive distribution of chickenpox incidence, which facilitates the computation of point forecasts and prediction intervals.

#### 1. Introduction

Chickenpox is a highly contagious disease caused by the varicella-zoster virus. Its main symptoms are a blister-like rash, itching, tiredness, and fever. It is a rarely fatal disease that mainly affects children younger than ten years of age, although older children and adults can also get it. Most of the people suffer the disease during childhood and reinfection is very strange [1]. Because of the contagious behavior of the disease, which spreads easily through the coughs and sneezes of an infected person or by touching the virus particles that come from chickenpox blisters, the best modeling approach would be obtained by taking into account individual movement and contact behavior. However, information at individual level is scarcely ever available and models based on aggregated data are usually considered.

Compartmental models are commonly used in epidemiology to understand the underlying mechanisms that influence disease transmission when aggregated data are available. These models divide the population being studied into different compartments according to the disease status and describe the evolution of the disease through changes in the number of individuals in each compartment. There may be susceptible, exposed, infectious, recovered, and immune individuals depending on the nature of the disease. Compartmental models are usually written using ordinary differential equations or difference equations. In addition, they are normally defined assuming that all the individuals in the population are equally likely to contact any other individual in that population [2, 3]. Models based on difference equations have been considered to describe the progression of chickenpox disease (see, e.g, [4, 5]). However, contact patterns in real populations are indeed heterogeneous. Therefore, models assuming homogeneous mixing should be replaced by models incorporating stochastic effects [2]. Stochastic models are able to accommodate the stochasticity inherent in the transmission of infectious diseases and provide an improved and more natural description of disease dynamics [6–8]. In addition, stochastic models can be analysed from a Bayesian viewpoint (see, e.g., [9]).

In the case of chickenpox, symptoms appear about two weeks (from ten to twenty-one days) after exposure to a contagious person. A person with chickenpox can spread the disease from one to two days before they get the rash until all the blisters have fully crusted over. The condition usually resolves by itself within a week and the infected individual usually acquires permanent immunity [1]. Therefore, a susceptible-exposed-infectious-recovered (SEIR) model seems an appropriate model to describe chickenpox dynamics.

In this paper, we present a Bayesian stochastic SEIR model in discrete time to understand chickenpox dynamics in the Valencian Community, Spain. The proposed compartmental model takes into account both heterogeneous mixing of individuals in the population and the inherent stochasticity in the transmission of the disease. In addition, the discrete formulation of the model better adjusts to the available data (weekly counts) and simplifies its Bayesian analysis. The analysis of the model from a Bayesian viewpoint allows us to consider a probability distribution for the observed counts of disease as well as for the parameters of the model. The posterior distribution for these parameters and the predictive distribution for future observations can then be straightforwardly obtained. Note that point forecasts and prediction intervals can be of utter importance to implement effective public health measures to reduce the burden of the disease.

The remainder of this paper is organized as follows. In Section 2 we present the available chickenpox data. Section 3 describes the SEIR model and shows its Bayesian analysis. In Section 4 we show the results obtained in the analysis of the data. Finally, we conclude with a general discussion of the proposed model and provide directions for further research.

#### 2. Data

In this paper we analyse weekly cases of chickenpox reported by the health sentinel network in the Valencian Community, Spain, for the period 2007–2012 (http://www.sp.san.gva.es). It is important to point out that there were no changes in the vaccination strategy during the period analysed. Starting in 2006, the chickenpox vaccine was introduced in the routine immunization schedule of the Valencian Community and, during the period 2007–2012, 10–14 year-old susceptible children were vaccinated. Because varicella vaccine was available in pharmacies, younger children could also be vaccinated against varicella beyond official guidelines.

For 2007–2012, an average of 19,000 cases of chickenpox have been reported per year in the region of Valencia, representing an average incidence rate of around 375 cases per 100,000 inhabitants. Figure 3 shows the time plot of the series, which presents a clear seasonal pattern. As can be seen, chickenpox incidence reaches the highest rates during spring. From June the trend is descending, with very few cases during the summer and part of autumn. In the end of autumn, we can observe an upward trend that continues during the winter to complete the seasonal cycle.

Chickenpox is a mandatory notifiable disease in the Valencian Community. On the other hand, population coverage by the Spanish Public Health System is almost universal (99.5%) and even persons not covered by it can also be attended to in public centers. Therefore, underreported chickenpox cases are assumed to be insignificant.

#### 3. Model

Although it can vary from person to person, an individual takes around two weeks to show symptoms after being infected and, as we mentioned before, a person with chickenpox can spread the disease from one to two days before the rash appears until all the blisters have crusted over (usually five to six days after the start of the rash). Hence, in order to describe as accurately as possible the different stages of chickenpox while keeping a relatively simple model, we make the following assumptions: once an individual comes into contact with a person with chickenpox, there is an incubation period of two weeks and then the individual becomes infectious for a week. Even though there is a proportion of infected individuals that, after being recovered, become susceptible again, that proportion is so small that it will not be considered here. Moreover, reinfected individuals do not transmit the disease. Consequently, chickenpox dynamics can be explained via a compartmental model that divides the population into susceptible, exposed, infectious, and recovered individuals. In addition, this model should focus on the first movement, that is, the transition from the susceptible compartment to the exposed one. Note that the following transitions (to the infectious compartment and finally to the recovered one) come naturally due to the nature of the disease.

Let be the number of people at the first week of their incubation period. Because the data reported by the health sentinel network refer to the number of infectious people at each week, represented by , it is necessary to transform the available data into the time-series . This can be easily done taking into account the fact that was . At the first level of the model hierarchy, we can assume that follows a Binomial distribution:where is the number of susceptible individuals at week and is the probability of getting the virus at week . In order to take into account the transmissible nature of the disease, we assume that the probability of infection depends on the number of infectious individuals at the previous week. Besides, this probability must lie in the intervals 0-1. In order to avoid the problem of estimates of falling outside the 0-1 limits, we model instead, which lies in the interval . In particular, we assume thatwhich is equivalent towhere is the transmission rate and the mixing parameter in (2) allows for heterogeneous mixing of individuals in the population [10]. Setting would correspond to the assumption of mass-action law. We assume here that follows the uniform distribution in the interval . A key feature of our model formulation is that the transmission rate is allowed to vary stochastically over time. In particular, we model as follows:where represents the number of sine-cosine waves needed to capture the seasonal variation in the disease transmission. Its value depends on the data under study, and so has to be estimated together with the other parameters of the model. Because the seasonal pattern varies slightly from year to year, the random effect allows for unspecified features of week . The parameters are assumed to have zero mean Gaussian distributions with standard deviations . The uniform distribution in the interval (0, 5) is assigned to all the standard deviations [11].

The number of individuals in each compartment is examined at discrete time steps through the following recursive equations:where is the number of susceptible individuals, is the number of individuals at the first week of their incubation period (which represents the data at hand, and so it is just set equal to its value), is the number of individuals at the second week of their incubation period, is the number of infectious individuals, and is the number of recovered (immune) individuals at week . Although the common formulation of the SEIR model includes only one compartment of exposed individuals, we divide here the compartment into two classes, and , in order to simplify the update of the recursive equations. By using this formulation, all the individuals in compartment at week move to compartment at week and so on. So the only compartments where individuals can stay longer than one unit time are the susceptible and recovered compartments. The average proportion of individuals vaccinated each week is represented by . The parameter represents the number of births at week and the population size, which is assumed to be constant over time (). To keep the population size constant, individuals must leave the system each week, proportionally to the sizes of the compartments; that is, . The flowchart diagram for the model is described in Figure 1.