Mathematical modeling of adulthood obesity epidemic in Spain using deterministic, frequentist and Bayesian approaches
Introduction
Obesity can be considered as a worldwide epidemic [1]. In Spain, the last survey from the Spanish National Health Survey (ENSE) shows that obesity has been increasing at worrying rate in the last decades for both infants and adults [2]. Obesity is not only a serious health concern, but also a public economic problem for the country [3]. Hence effective policies and control measures are necessary in order to mitigate the augment of excess weight in Spain.
Mathematical models are a tool to analyze the evolution of the epidemic, its future dynamics, manage and control its increase and prioritize policies that permit preventing and reducing or at least alleviating the spread. Most of the approaches in mathematical modeling are based on using compartmental models [4], [5], [6], [7]. The population is divided into compartments according to the status and the individuals flow between the compartments as time passes. The mathematical model is usually based on systems of ordinary differential equations in which homogeneous mixing is assumed.
These types of models have been applied to explain the evolution of excess weight in Spain. In the case of the region of Valencia, Spain, we find two models: one for infants [8] and another one for adults [9]. They are based upon systems of ordinary differential equations with input parameters, where the Valencian population is divided into compartments according to weight. Persons move between the compartments as time goes by. Of course, many factors may play a role in body weight: genes, metabolism, environment, culture, etc. In [8], [9], the authors focused on the fact that fatness is “transmitted” by social pressure and contacts that entail unhealthy habits (this includes social environment, media, culture, access to food, economical status, etc.). Overweight and obese individuals are viewed as infected, which transmit the disease to healthy persons (susceptible) by contact. This idea was developed in [10], [11]. In the seminal paper [10], the authors evaluated a densely interconnected social network to analyze the extent of the person-to-person spread of obesity as a possible factor contributing to the obesity epidemic; they explicitly concluded that obesity appears to spread through social ties. Paper [11] and its abundant references from the introduction defend the idea that gains in weight are spread through the population when social surrounding influences healthy weight individuals, in a way reminiscent of a contagious disease of social transmission. In their words, it may be easier to be fat in a society that is fat. Hence the transition from healthy weight to overweight in a certain moment of one’s life may be justified by his/her contacts in the society with overweight and obese persons (the infectives).
This paper proposes a mathematical model for the number of overweight and obese adults in Spain as a disease of social transmission, by taking into account data gathered by the Spanish National Health Survey from 1987 to 2017. In general, finding abundant reliable data is difficult because of the hard preparation and storage and the economic cost. We use a nonautonomous nonlinear system of ordinary differential equations, whose parameters are estimated by minimizing the mean square error. Accounting for modeling and data measurement errors, randomness is incorporated into the model, and frequentist nonlinear regression and Bayesian inference are carried out [12]. A sensitivity analysis is conducted for designing the optimal strategy for reducing the excess weight prevalence in Spain. In epidemiology, the quantification of the propagation of input uncertainties has been investigated in [13], for some prototypical continuous compartmental models, in [14], for discrete compartmental models, and in [15], for the obesity model proposed in [9], by utilizing generalized polynomial chaos expansions [16], [17], but no inverse parameter estimation method. In our recent work [18], we inferred consistent probability distributions for the input parameters of the obesity model proposed in [9] according to their deterministic estimates, by employing the maximum entropy principle. The use of the maximum entropy principle may be justified in the following situations: there are replicated data (which is not the case for an epidemic scenario); or the amount of data at hand is very limited (like four instants of time or less) and the model parameters are therefore fixed empirically or from earlier technical reports. On the other hand, Bayesian inference has been carried out for the analysis of epidemics such as respiratory syncytial virus and chickenpox infections in the region of Valencia [19], [20].
The structure of this paper is as follows. In Section 2, we explain the study from the Spanish National Health Survey and present the data that will be used. In Section 3, we formulate the deterministic mathematical model for adulthood excess weight in Spain, we fit the parameters, study the dynamics and conduct a sensibility analysis. Section 4 adds a random white noise error to the governing model accounting for model errors, with constant parameters, and frequentist nonlinear regression is carried out to estimate the error variance and to construct prediction intervals for the incidence of obesity. In Section 5, we work with random input parameters to apply the Bayesian methodology, accounting for model and data errors at the same time. Different versions of the Metropolis algorithm (namely standard and Adaptive) are implemented to simulate from the posterior distribution and the posterior predictive distribution. Finally, Section 6 draws the main conclusions and points towards potential avenues of future research.
The code implementations and the numerical computations have been developed in the software MathematicaⓇ, version 11.2 [21].
Section snippets
Survey and data
The Spanish National Health Survey (ENSE) is a periodic study conducted by the Spanish Ministry of Health, Consumption and Social Welfare, with the collaboration of the National Statistics Institute of Spain (INE), since 1987. It is a massive scale survey based on home personal interview (CAPI) that collects transversal data regarding the resident population in Spain, on aspects relating to health information, namely medical conditions, use of health services, and health determinants. It makes
Deterministic model
In this section we propose an epidemic-based mathematical model for the data from Fig. 1. The construction is based on a system of ordinary differential equations, where the time-dependent variables refer to the overweight and obese individuals in Spain who are at least 18 years old.
Frequentist-based model
In mathematical modeling, there are three sources of error: model error, numerical errors, and data errors. The first one is due to the incomplete knowledge of the underlying physics. The second one comes from the resolution of differential equations, optimization problems, etc. Finally, the third one accounts for error measurements. Numerical errors can be reduced as much as desired, at least theoretically, by employing finer discretizations, better algorithms, etc. However, model error and
Bayesian model
As explained at the beginning of Section 4, randomness should be included into the model accounting for modeling and data errors. In Section 4, randomness was incorporated through an error term of white noise form, keeping the input parameters as constants, accounting for model errors. The validation of the model output consisted of a pointwise estimation, which coincided with the deterministic estimation from Section 3, and a prediction interval that captured all the actual data. In this
Discussion and perspectives
In this paper, we have modeled successfully the number of excess weight adults in Spain for the last thirty years. We have started with a well-posed deterministic model of ordinary differential equations, in which excess weight is treated as an epidemic where the individuals flow between different compartments according to their fatness. The parameters of the model have been estimated by minimizing the objective function defined through the mean square error. Accounting for model and data
CRediT authorship contribution statement
Julia Calatayud: Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Supervision. Marc Jornet: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Supervision.
Declaration of Competing Interest
The authors declare that there is no conflict of interests regarding the publication of this article.
References (42)
- et al.
A deterministic model for highly contagious diseases: the case of varicella
Physica A
(2016) - et al.
Modeling dynamics of infant obesity in the region of Valencia
Comput Math Appl
(2008) - et al.
Computational uncertainty quantification for random time-discrete epidemiological models using adaptive gPC
Math Method Appl Sci
(2018) - et al.
Linear models in statistics. 2nd edition
(2008) - et al.
A stochastic collocation approach to Bayesian inference in inverse problems
Commun Comput Phys
(2009) Health communication campaigns and their impact on behavior
J Nutr Educ Behav
(2007)- et al.
The worldwide obesity epidemic
Obesity
(2001) - Spanish national health survey 2017 (ENSE 2017). 2019. Spanish Ministry of Health, Consumption and Social Welfare,...
Costes sociales y económicos de la obesidad y de sus patologías asociadas
(1999)Compartmental models in epidemiology
Mathematical epidemiology
(2008)
Dynamical modeling and analysis of epidemics
Mathematical modelling of respiratory syncytial virus (RSV): vaccination strategies and budget applications
Epidemiol Infect
Mathematical modelling of social obesity epidemic in the region of Valencia
Spain Math Comp Model Dyn
The spread of obesity in a large social network over 32 years
N Engl J Med
Imitative obesity and relative utility
J Eur Econ Assoc
Uncertainty quantification: theory, implementation, and applications
Epidemic models with random coefficients
Math Comput Model
Uncertainty quantification in simulations of epidemics using polynomial chaos
Comput Math Method M
Numerical methods for stochastic computations
A spectral method approach. new york: cambridge texts in applied mathematics
The Wiener-Askey polynomial chaos for stochastic differential equations
SIAM J Sci Comput
Stochastic simulation of epidemics using the maximum entropy principle and generalized polynomial chaos expansions
Rom Rep Phys
Cited by (9)
Prevalence of obesity and associated cardiovascular risk factors in the Spanish population: the ENPE study
2022, Revista Espanola de CardiologiaModeling of adulthood obesity in Spain using Itô-type stochastic differential equations
2021, Chaos, Solitons and FractalsCitation Excerpt :See [18,25]. In this paper, the deterministic compartmental model from [18] is randomized in a different manner, via noise. The derivatives of the compartments are perturbed by Gaussian white noise-type pure errors that have a rough and unpredictable structure.
Power-series solutions of fractional-order compartmental models
2024, Computational and Applied MathematicsA dynamical mathematical model for crime evolution based on a compartmental system with interactions
2024, International Journal of Computer MathematicsTheory and methods for random differential equations: a survey
2023, SeMA JournalSpatio-temporal stochastic differential equations for crime incidence modeling
2023, Stochastic Environmental Research and Risk Assessment