OBESITY EPIDEMICS MODELLING BY USING INTELLIGENT AGENTS

Transcription

1 OBESITY EPIDEMICS MODELLING BY USING INTELLIGENT AGENTS Agostino G. Bruzzone, MISS DIPTEM University of Genoa Eail - URL t Vera Novak BIDMC, Harvard Medical School Eail vnovak@bidc.harvard.edu - URL Francesca Madeo Siulation Tea Genoa University Eailadeo@siulationtea.co - URL Introduction The obesity epideic is a huge proble spreading worldwide in the past three decades (Wolf et al. 27). The obesity is an hard proble in ter of people healthy but also in ter of social cost. In fact it was deonstrated that the obese individuals ipact on national health care syste costs exceeds the average expenditure per capita of a noral weight individual; therefore uch of the social costs associated with obesity are due to different social interactions that obese individuals ay have, always considered in coparison to a noral weight individual (Cereda 211). A recent study in Australia has estiated the intangible costs arising fro obesity in adulthood, reaching a value between 13 to 18 billion AU$; in addition to these aspects, any studies have proposed estiations of indirect costs of obesity highlighting the fact that their econoic ipact could exceed the direct health care costs, both in ter of absolute and percentage of GDP (Magarey et al 21). For instance in the U.S.A., it is estiated a loss of production due to the obesity corresponding to 23 billion USD ( ), while in Australia the loss for the country is estiated into 272 illion AU$ (Australian Bureau of Statistics). The Siulation Tea Genoa University is cooperating with the Harvard Medical School in order to provide M&S (Modeling and Siulation) tools for odeling huan behavior and its influence on the evolution of obesity epideics and its effects on social networks, infrastructures and facilities. In particular the authors propose and agent based siulation approach, due to the fact that It is evident that in health care ore than in other sectors the high influence of stochastic factors and very coplex correlation ake difficult analysis on large scale without using odeling and siulation; so a very interesting aspect is to obtain a better understanding of the phenoena generating/affecting current developent of obesity epideics that is dynaically and quickly evolving in any countries; and therefore, providing challenges today and threats for the future. In particular, the Siulation Tea proposes an Agent Based Siulation Approach for odeling huan behavior and its influence on the evolution of obesity epideics and its effects on social networks, infrastructures and facilities in order to provide and to provide support for decision akers. A first iportant benefit expected by these siulators should be the possibility to test and to validate the different existing hypotheses related to the utual influence aong different factors related to physiological and psychological issues, behavioral aspects and regional/ethnical/social/geographical and econoical factors. In fact, by conducting experients using inforation available in larger saples, it will allow to validate the consistency of these predictions within a population over tie and also across diverse populations. Once the odels becoe validated, it will be possible to use a forecasting syste, that would allow to conduct risk analysis and estiation of future developent; the siulator, used in this way, reproduces scenarios and generates forecasts in ter of resources and facilities required for treatents, as well as estiation of the ipact of costs and deand on the country infrastructures Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 18

2 and industry. It would also allow to predicting the ipact of behavioral, social and econoic interventions to prevent further developent of obesity epideics and to curtail its costs. Finally such siulators could becoe a very strategic advantage to support decision and to evaluate the ipact of actions and countereasures of public and private institutions and organizations on such a critical sector as the health care. The authors decided to start a joint research on these issues by focusing on obesity due to its strong ipact on country econoies and to its very coplex and dynaic evolution; in fact obesity of a population evolve based on individual and social behaviors over tie (i.e. depression due to soe social probles, lack of obility, socially acceptable overeating etc. ). The idea to reproduce the phenoena related to obesity is to use intelligent agent reproducing the population and let the evolve based on their behavior and on the applied actions and scenario evolution; the behavioral odels are defined inside the agents, defined IA-CGF (Intelligent Agent Coputer Generated Forces); for instance an high level exaple is proposed in figure 1; a cobined stochastic siulation engine anage the interactions aong the agents; this fraework defined as NCF (Non Conventional Fraeworks) could be tailored over very different contexts and areas. Previous researches have been carried out by Siulation tea to develop odels able to reproduce huan behavior over town or regions; in fact these odels were used originally for epideic evolution, for analyzing urban disorders and for country reconstruction. In this case it was decided to create an ad hoc NCF with full inheritance of IA-CGF Libraries. For the specific research it was possible to start the developent of new conceptual odel and to design a first shell of NCF defined BACCUS (Behavioral Advanced Characters and Coplex Syste Unified Siulator) that introduce several additional paraeters related to physiology, health status and behavior. Obesity Epideics The obesity epideic affects people of all ages including children, both genders, different ethnic and racial backgrounds, and various socioeconoic groups. Adult overweight and obesity are defined using body ass index (BMI): noral weight < 25; overweight: ; obesity > 3 W BMI = 2 H BMI W H Body Mass Index Weight [kg] Height [] Children at risk for overweight are defined as the 85th and 95th percentiles of body ass index. Currently, (21), no state has prevalence of obesity less than 2% and 26 states had a prevalence of 25% or ore (Centers for Disease and prevention; 66% of adults are overweight or obese; 16% of children and adolescents are overweight, and 34% are at risk of overweight. Minority and low-socioeconoicstatus groups are disproportionately affected at all ages. By 215, 75% of adults will be overweight or obese, and 41% will be obese. Obesity is associated with increased risk of diabetes, hypertension, cardiovascular diseases, strokes and deentia, obility dysfunction, cancer and ortality thus increasing significantly health care cost in the society. Elevated body ass index is being increasingly recognized as a risk factor for stroke, cardiovascular disease, and cognitive decline (Falkstedt et. al. 26; Cournot et. al. 26). Diabetes epideic follows obesity spread in the world, affecting countries that previous had a lower prevalence of diabetes. It is expected that diabetes prevalence of people with diabetes will draatically increase further between 21 and 23 i.e. India fro 5.8 to 87 illions; China fro 43.2 to 62.6;United States of Aerica fro 26.8 to 36.;Pakistan fro 7.1. to 13.6; Brazil fro 7.6 to 12.7 (Wang et al. 27). A long-ter population study with 27 years of the follow-up has shown prospectively that in the ultiethnic population, idlife obesity increases the risk of deentia later in life. Obese people (BMI 3) had a 74% increased risk of deentia (hazard ratio 1.74, 95% confidence interval 1.34 to 2.26), while overweight people (body ass index ) had a 35% greater risk of deentia (1.35, 1.14 to 1.6) as copared with those of noral weight (body ass index ) (Whiter et al. 25). The increased risk for Alzheier s disease and deentia later in life is independent even if adjusted for elevated blood pressure, soking, socioeconoic status and genetic factors (Wolf et al.27;kivipelto et al.25). Obesity epideic is associated with significant burden for the people, failies and society and Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 19

3 contributes significantly to increased health care cost, orbidity and ortality. The exact cost of obesity to the society and people is not known. However, a cost of diabetes could be used as an exaple; i.e. in billions of international dollars, 8787 ID per person in the world and 55.7 billions of national incoe loss in China alone. It is expected that diabetes will increase death rates globally by 17% in 23 and by 25-27% in Middle East, India and China (IDF Atlas 4 th ed. International Federation on Diabetes). Therefore, health care cost will increase exponentially, as the obesity epideics spreads into younger population and other countries. Recently soe researchers have carried out a study with respect to the Chinese context, the analysis was derived fro two reference years, 2 and 225 (Popkin et al 26); based on these estiates it results, for the year 2, an econoic ipact of 49 billion USD (4.6% of GDP), of which 43 billion dollars in indirect costs (3.58% of GDP). The agnitude of the econoic ipact is expected to reach a size even ore critical, both for increased health spending and the consequences on the labor arket. The projections to the year 225 describe an expected total cost overpassing 112 billion USD representing 9.23% of GDP, of which 16 billion USD (8.73% of GDP) is attributable to indirect costs; these researches estiated that, even in China, the largest coponent of costs was represented by the loss of productivity due to absences due to illness, which causes 75% of indirect costs. The issues related to sickness, early retireent and disability, have been investigated in Sweden, Finland and Denark (World Health Organization 1997). The results produced have highlighted the link between increased BMI and sickness absence in the long ter. Furtherore, considering the child obesity as a risk factor for obesity in adulthood, a fraction of the costs, direct and indirect, previously entioned, the adult is generated fro the high nuber of adolescents in which obesity has persisted over tie. In this direction, soe researchers have investigated the effects expected along the course of life resulting fro a weight reduction progra ipleented in Aerican schools. It was introduced the paraeter QALYs to quantify the results obtained: the QALY (Quality Adjusted Life Years acrony) is a unit of easure used in cost utility that cobines the life span with the sae quality. One QALY equal to 1 corresponds to the expected life of one year in noral health, the value corresponds to death (Pliskin et al. 198). The easureent scale is continuous and a few years of life ay also be given negative values (if you have serious conditions and acute suffering of iobility). QALY is used as an index weighting in the assessent of increases in life expectancy associated with health interventions. Thus, for exaple, whether the introduction of a new surgical technique allows the patient to survive on average 6 years older, but the conditions are such that after the operation be considered equal to.2 QALY (eg., Because of serious otor deficiencies and frequent pain), the intervention effect on life quality will be weighted for only 1.2 years. Cost-utility analysis perfored in the above studies was a cost of 435 USD/patient per QALY gained, while the su of the avoided health care costs and productivity losses avoided, you get an expected benefit, net of ipleentation costs of 7313 USD/patient for each progra ipleented (Wang et al.23). Obesity is a ultifactorial process that results fro interactions aong the individual health status, functional and social habits, social networks, education and other factors that cannot be predicted fro a single variable (i.e. body ass) but requires nonlinear odeling of ultiple variables and their interactions. Although that reduced physical activity, increased food intake and social networking are coonly cited factors, the echanis underlying obesity spread worldwide echaniss reain poorly understood. The echaniss of obesity spread and their ipact on the world s health, personal life and econoy have not been well studied and therefore, the effective strategies to prevent obesity epideics are lacking. Modelling Obesity Epideics The idea to reproduce the phenoena related to obesity is to use intelligent agent reproducing the population and let the evolve based on their behavior and on the applied actions and scenario evolution (Bruzzone et al.211); the behavioral odels are defined inside the agents, defined IA- CGF (Intelligent Agent Coputer Generated Forces); for instance an high level exaple is proposed in figure 2; a cobined stochastic siulation engine anage the interactions aong Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 2

4 the agents; this fraework defined as NCF (Non Conventional Fraeworks) could be tailored over very different contexts and areas. Previous researches have been carried out by Siulation tea to develop odels able to reproduce huan behavior over town or regions (Bruzzone et al.28); in fact these odels were used originally for epideic evolution (Avalle et al.1999), for analyzing urban disorders (Bruzzone et al.26) and for country reconstruction (Bruzzone & Massei 21). In this case it was decided to create an ad hoc NCF with full inheritance of IA-CGF Libraries and it was developed a first shell of NCF defined BACCUS (Behavioral Advanced Characters and Coplex Syste Unified Siulator) that introduces several additional paraeters related to physiology, health status and behavior; these paraeters includes: BMI Sport Profile Alcohol Profile Stroke Infarct Diabetes Cancer Hypertension Atrial Fibrillation Hyperlipideia The BACCUS siulator is ipleented using IA- CGF engine of the Siulation Tea as anticipated and it is proposed in Figure 3. The developent concept is based on a several step process; first phase is related to defining the population in ter of PIG (people initialization groups); the PIGs represent the different groups present on a scenario; PIGs are defined in ter of statistical distribution for their ain factors (Social Level, Educational Level, Political Attitudes, Religion, Ethnic, Tribe, Gender, Age); based on these paraeters the population is generated randoly respecting original statistics on the geographic region, but even considering the different groups characteristics with their specific structure (i.e. soe ethnic group have different social status statistics respect other one living in the sae region); this process generates the people agents; in addition inside the region it is possible to define Zones as objects that have affinities with the ain factors and so the people agents are distributed geographically in ter of living and working locations in consistency with Figure 4. Investigation Process of the Obesity over a Region based on Modeling the Population Behavior their affinities; it is possible to overlap zones with different affinities in order to represent inconsistent ixes of different groups of the population. In addition to the individual aspects, in the odel the aggregation paraeters are defined to regulate the generation of social networks based on stochastic distributions; by Montecarlo technique the people agents are associated in ter of failies and working connection generating the social network; the behavior of each agent is defined by odels that regulates how it operates under regular (i.e. working days, holidays) or special conditions (i.e. natural disasters, sickness period); an overall presentation of the procedure is proposed in Figure 4. Obviously dealing with huan odeling the verification and validation of the siulator is critical as well as the data collection and analysis; for the preliinary test the hypotheses on the conceptual odel relating behaviors with obesity was based on current researches (Christakis et al. 27) as well on the analysis of data available in Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 21

5 1 9 Beth Israel Deaconess Medical Center (BIDMC) in Boston and Harvard Medical School affiliate. In fact soe general data used for siulation were derived fro previous studies obtained by the databases of associations (i.e. World Health Organization, the Aerican Heart Association and the Aerican Diabetes Association). Table I: Extract of Results based on non paraetric rank correlation on available saples Table II: Soe Correlation Extract aong factors and BMI based on the available saples these 17 patients in the saple was volunteers, aged between 5 and 85 years, recruited fro the SAFE Lab for four saples, conducted in the years 26 to 21. All four saples considered issues related to diabetes; in fact the data were related to studies correlating diabetes with other pathologies, so the saple includes other diseases such as patients with yocardial infarction or stroke. All saples have a quite balanced ratio between balance between ale patients and feale patients. There are soe liitations in using saples of existing populations that were collected or originated for different studies, and therefore do not provided sufficient inforation i.e. about frequency of easureents, include specific populations, or have issing values for certain paraeters. Therefore, this paper represents a first step forward for odeling and tuning siulators for investigating obesity, so the focus is ostly on creating the conceptual odels, introducing consistent data and tuning the paraeters in order to obtain reasonable results by the siulation runs. Soe exaples, of the paraeters used to check utual influences are reported in table I e II; statistical analysis and ranking ethodologies was applied in order to check data significance and their correlation. The low level of correlation obtained was expected due to the reason above entioned; therefore the analysis was useful to identify procedures and aspects to be investigated in future data collections and researches. In order to proceed it was decided to ipleent soe behavioral odel and correlation algoriths aong the factors based on author's hypotheses consistent with the data available. Mean Square pure Error Regular OverWeight Obese BMI [kg2/4] Figure 5. IA-CGF in BACCUS Siulator allows to evaluate the population behavior and its influence on the obesity evolution in the town In addition it was used a saple related to 17 subjects in the Syncope and Falls in the Elderly Laboratory (Lab SAFE), active in the BIDMC; Y (MSpE) Tie [days] Figure 6. BACCUS VV&A based on Dynaic Analysis of Mean Square pure Error on Population Obesity Classes Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 22

6 The scenario used to the tests was a sall town of about 15' inhabitants; target functions included the population sharing aong the different obesity classes and the average BMI: Regular Overweight Obese Average BMI In figure 5 it is proposed the BACCUS during execution of the proposed scenario, analyzing each single individual. The VV&A (Verification, Validation & Accreditation) of BACCUS is based analysis of MSpE (Mean Square pure analysis) as easure of the variance of the target functions aong replicated runs over the sae boundary conditions; by this approach it becoes possible to identify the nuber of replications and the siulation duration able to guarantee a desired level of precision; MSpE values in correspondence of these experiental paraeters deterines the aplitude of the related confidence band: MSpE CBA t the no Srk k(t) tie of ( t, n ) 1 () t Sr () n t j ( t, n, ) = t MSpE ( t, n ) = n n Sri i= 1 j= 1 α ± α, n n siulation tie -th target function estiated by siulator nuber of replications with sae boundary conditions and different rando seeds -th target function value at t the k-th replicated siulation MspE (t,n o, α) Mean Square pure Error at t tie and with no replications for the -th target function α percentile CBA (t,n o, α) Confidence Band Aplitude at t tie, with no Replications for the -th target function In fact the MSpE allows to quantify the experiental error due to influence of the stochastic coponents; as presented in figure 6 2 the variance of all the target function reach steady state situation over a reasonable nuber of replications and over a tie horizon of about 1 year. so this confir that siulator provides consistent results on a stable situation with capability to define the confidence band for estiating the obesity target functions. Conclusion The use of agents and siulation to investigate large scale health care probles represent an iportant opportunity; obviously in these cases it is critical to guarantee a ultidisciplinary approach to the proble. The obesity epideic represents a very iportant and interesting application fraework that could be very useful to consolidate research in this area of M&S related to Medicine and Health Care. The research highlighted the critical aspects related to collecting, ining and filtering the data to define the conceptual odels related to such coplex probles as well as to support paraeter fine tuning and siulator VV&A. The odel and the present results in this first phase are proising and the potential of using Intelligent Agents in this context is very great considering the ipact of the obesity epideic. References [1] Bruzzone A.G. Treori A., Massei M., Adding Sart to the Mix, Modeling Siulation & Training: The International Defense Training Journal, 3, 25-27, 211 [2] Cereda C., Models and Analysis of Coplex Systes for the Evaluation of Future Scenarios and of Related Infrastructural and Operational Needs, Genoa University Thesis, DIPTEM Press, 211 [3] Bruzzone A.G., Massei M., Intelligent Agents for Modelling Country Reconstruction Operation", Proceedings of AfricaMS 21, Gaborone, Botswana, Septeber 6-8, 21 [4] Bruzzone A.G., Scavotti A., Massei M., Treori A., Metaodelling for Analyzing Scenarios of Urban Crisis and Area Stabilization by Applying Intelligent Agents, Proceedings of EMSS28, Septeber 17-19, 28, Capora San Giovanni (CS),Italy, 28 [5] Wolf PA, Beiser A, Elias MF, Au R, Vasan RS, Seshadri S: Relation of obesity to cognitive function: iportance of central Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 23

7 obesity and synergistic influence of concoitant hypertension. The Fraingha Heart Study. Curr Alzheier Res 4: , 27 [6] Wang Y, Beydoun MA: The Obesity Epideic in the United State- Gender, Age Socioeconoic, Racial/Ethnic and Geographcis Characteristics : A systeatic Review and Meta-Regression Analysis. Epideiologic Reviews 29:6-28, 27 [7] International Journal of Obesity 31: , 26 Neurology 67: , 26 P [8] Popkin B.M., Ki S., Rusev E.R., Du S., Zizza C., Measuring the full econoic costs of diet, physical activity and obesity-related chronic diseases, Obesity Review, 7, , 26 [9] Bruzzone A.G., Bocca E., Rocca A., Algoriths devoted to Reproduce Huan Modifiers in Polyfunctional Agents, Proc. of SCSC26, Calgary, Canada, July 3- August, 26 [1] Whiter RA, Gunderson EP, Barrett- Connor E, Quesenberry CP, Jr., Yaffe K: Obesity in iddle age and future risk of deentia: a 27 year longitudinal population based study. BMJ 33:136, 25 [11] Kivipelto M, Ngandu T, Fratiglioni L, Viitanen M, Kareholt I, Winblad B, Helkala EL, Tuoilehto J, Soininen H, Nissinen A: Obesity and vascular risk factors at idlife and the risk of deentia and Alzeheier disease. Arch Neurol 62: , 25 [12] Australian Bureau of Statistics, National Health Survey 24-5: Suary of results. ABS cat.no Canberra, 25 [13] Wang L.Y., Yang Q., Lowry R. and Wechsler H.,Econoic Analysis of a School-Based Obesity Prevention Progra, Obesity Research 11, , 23 [14] Magarey AM, Daniels LA & Boulton JC, Prevalence of overweight and obesity in Australian children and adolescents: reassessent of 1985 and 1995 data against new standard international definitions. Medical Journal of Australia 174: , 21 [15] Avalle L, A.G. Bruzzone, F. Copello, A. Guerci, P.Bartoletti Epideic Diffusion Siulation Relative to Moveents of a Population that Acts on the Territory: Bio- Dynaic Coents and Evaluations, Proc. of WMC99, San Francisco, January, 1999 [16] Pliskin, J., Shepard, D, Weinstein, M, Utility Functions for Life Years and Health Status". Operations Research (Operations Research Society of Aerica) 28 (1): , 198 [17] Spearan C.E. "The Abilities of Man: Their Nature and Measureent", Macillian, London, 1927 [18] Foster G.D., What is a reasonable weight loss? Patients' Expectations and Evaluation of Obesity Treatent Outcoes, Journal of Consulting and Clinical Psychology, 65 (1): 79-85, 1997 [19] WHO, World Health Statistics: Reports, 211 [2] WHO, Prevening and Managing the Global Epideic of Obesity: Report of the World Health Organization Consultation of Obesity, Geneva, June 1997 [21] National Institutes of Health, Clinical Guidelines on the identification, Evaluation and Treatent of overweight and Obesity in Adults: The Evidence Report, 1998 Bruzzone, Novak and Madeo SCS M&S Magazine 212 / n.1 24