Methodology and Modeling Environment for Simulating National Health Care

Methodology and Modeling Environment for Simulating National Health Care Bernard P. Zeigler, Ernest Carter, Chungman Seo, Cynthia K Russell*, and Brenda A. Leath* RTSync Corp. 530 Bartow Drive Suite A Sierra Vista, AZ, 85635, USA zeigler@rtsync.com, earnest.carter@rtsync.com, cseo@rtsync.com Keywords: DEVS M&S, Health Care System, Systems of Systems (SoS), Agent Based Modeling, System Entity Structure (SES) Abstract An ideal health care system is unlike today s fragmented, loosely coupled, and uncoordinated assemblage of component systems. An ideal (optimal) health care delivery system will require methods to model large loosely coupled distributed system of systems. This paper presents a modeling and simulation methodology to support design of coordinated care architectures as well as to predict important quality versus cost metrics of such architectures. Such a model would be both useful in itself while illustrative of a general framework for system of system level modeling. An essential and challenging aspect of such modeling is to include models of human behavior to provide a veridical basis for quality and cost assessment of coordination architectures. 1. INTRODUCTION A recent workshop [1] envisioned an ideal health care system that is unlike today s fragmented, loosely coupled, and uncoordinated assemblage of component systems. The workshop listed one of the specific challenges to model large-scale, distributed systems where loose coupling occurs, commenting that, An ideal (optimal) health care delivery system will require methods to model large scale distributed complex systems. Such knowledge would evidently be useful for other industries as well. Improving the health care sector presents a challenge that has been identified under the rubric of Systems of Systems (SoS) engineering [2, 3], in that the optimization cannot be based on suboptimization of the component systems, but must be directed at the entire system itself. While current industrial and system engineering models and simulations have targeted the micro (component system) level, it is the macro (system of systems) level that such approaches must now address. While trying to model the entire health care system is overwhelming, restricted system of system level models can be developed to address specific *Westat 1600 Research Boulevard Rockville, MD 20850-3129 CynthiaRussell@westat.com, BrendaLeath@westat.com issues while still enabling a general framework for engineering of the overall system of systems to emerge. One such issue, the ability to examine the tradeoff between quality and cost of coordinated care is of intense interest to Accountable Care Organizations (ACO). According to the Centers for Medicare and Medicaid (CMS) web site, ACOs are groups of doctors, hospitals, and other health care providers, who come together voluntarily to give coordinated high quality care to the Medicare patients they serve. CMS is providing incentives to establish such ACOs through its offer to share any savings produced by an ACO in delivering service without sacrificing quality. However, currently there is no basis for negotiating with potential ACOs on the structuring of such profit sharing arrangements. An adequate basis for such negotiations would be an accepted means of measuring quality likely to be achieved by an ACO proposal versus the cost that would be involved. Such a basis is currently lacking. A system-level model that could reliably predict the quality versus cost performance of an ACO s proposal would provide a basis not only in negotiations between CMS and potential ACOs but also serve in numerous additional ways such as in developing and testing tools and services for ACOs. As central players in federal attempts to bend the cost curve, such organizations will become key to cutting costs while maintaining and even improving quality of delivered care. Health care spending in the United States is highly disproportionate, with half of health care dollars spent on five percent of the population. ACOs that manage this at risk population coordinate the care of patients who need special attention after leaving the doctor s office or the hospital because they cannot carry out the activities required to complete their care. People with multiple health and social needs are high consumers of health care services, and are thus drivers of high health care costs. The ability to provide the right information to the right people in real time requires a system-level model that identifies the various community partners involved and rigorously lays out how their interactions might be effectively coordinated to improve care for the neediest patients that cost the most. This paper presents a modeling and simulation methodology to support design of coordinated care

architectures employed by ACOs as well as to predict important quality versus cost metrics of such architectures. Such a model would be both useful in itself while illustrative of a general framework system of system level modeling. An essential and challenging aspect of such modeling is to include models of human limitations that coordinated care is intended to ameliorate. Modeling of human capabilities, especially intelligent ones, is typically referred to as agent-based. We discuss development of a SoS simulation model that takes account of emerging health information networks and electronic medical records, and integrates system concepts together with agent modeling concepts extended to include human behavioral limitations. This will provide a basis for developing and testing architectures for health care services with application to coordinated care systems as examples. We will follow the approach of Aumann [4] which is to determine and characterize the primary (focal level) of model development and those levels immediately above and below together with their experimental frames. The focal level is where the behaviors of the system of interest reside. The mechanism for a model behavior arises at the lower level while its purpose is found at the higher level. Defining the relationships among these levels will determine the means by which data and parameter values and constraints will flow between them. For example, if the focal level is the individual level, as in Agent-based models, then models and frames at the immediate levels above (e.g., society) and below (e.g., decision and action components) this level must be considered and related to those at the focal level [4]. The rest of this paper is organized as follows: Section 2 provides background in coordinated care, requirements for systems level integration and models of personal limitations. Section 3 presents a methodology and modeling environment for simulating national health care applicable to coordinated care architectures based on multi-level modeling and families of models. Section 4 provides conclusions on validation of the model for successful coordinated care program for women with high risk pregnancies as well as how the Discrete Event Systems Specification (DEVS) and SES formalisms support the modeling methodology 2. BACKGROUND: COORDINATED CARE It has been found that 15% of the most expensive health conditions account for 44% of total health care costs; patients with multiple chronic conditions cost up to seven times as much as patients with one [5-9]. To date there has been only a few studies of coordinated care showing promising results. A program run under the aegis of Rockville Institute s Pathways Community Coordination framework has been in operation in Ohio since 2005. In Ohio, low-weight births represent only about 10 percent of all Medicaid births but account for more than 50 percent of all Medicaid birth expenditures. The Community Pathways Model was developed by the Ohio Community Health Access Project (CHAP) in Richland County to improve health and preventative care for high-risk mothers and children in difficult-to-serve areas. The existence of such programs, and the prospect for many more being set up by ACOs, provides the motivation for model development. On the one hand, their success suggests that the approach should be replicated; on the other hand, there is not enough understanding about why these programs are successful and particularly, why their success is limited to only a portion, although a relatively large portion, of the population they address. A SoS simulation model can relate success in terms of outcomes in relation to resources and costs with the specific attention to the relevant characteristics of patients being treated. Coordinated care presupposes that continuing intervention after patients return home can greatly reduce unnecessary retreatment and declining health with the associated reduction in the extra burdens that such populations place on health care resources [10]. Falls and non-adherence to prescribed medical regimens are two areas which have been identified as sources of high cost [11, 12], which have not been controlled despite introduction of current assistive devices [13, 14]. The questions we are interested in addressing for the model to be developed are: a) how exactly can such care be provided in a coordinated manner and b) will the cost of such additional care be significantly less than the savings that are achieved by dedicating the resources needed to provide such care? 2.1. The Crucial Role of System-Level Integration Craig et al. [15] present a care coordination model aimed at improving care at lower cost for people with multiple health and social needs. The concept of model here is rather a framework for identifying the players and interactions involved in coordinated care and ferreting out the features of successful organizations. Within this framework, Craig concludes that Communities that have the most success in coordination efforts have skilled care managers implementing individual-level interventions as well as effective leadership aligning key stakeholders of multiple systems (e.g., county public health department, state mental health office, hospital administration, housing nonprofit leadership.) The ability to provide the right information to the right people in real time requires a system-level model

that identifies the various community partners involved and rigorously lays out how their interactions might be effectively coordinated to improve care for the neediest patients that cost the most. Although frameworks such as that of the Institute for Healthcare Improvement [15] provide a starting point, currently a rigorous predictive integration model is lacking. For example, the IHI framework does not take account of emerging health information networks and electronic medical records, nor does it integrate system concepts and agent modeling concepts extended to include human behavioral limitations. Consequently, it does not provide a basis for developing and testing architectures for health care services with application to coordinated care systems as examples. A SoS model for coordinated care would enable systems design, engineering, and evaluation of alternative coordinated care structures. Such a model would focus on information technology support and apply to a wide variety of structures of possible interest to ACOs, accounting for the populations that ACOs serve, and the community resources available in their environments. 2.2. Modeling and Simulation for Industrial, System, and Systems of System Engineering Modeling and Simulation in industrial and system engineering has a long history of application with an enormous literature, well represented by a classic text [16]. Unfortunately, application in health care has been primarily to micro-systems such as hospital resource allocation and training systems for surgery and other interventions. As indicated above, recently there has been recognition of the need for a concept of Systems of Systems (SoS) engineering [2], that is directed at the requirements for optimization at the macro level which is characterized as an assemblage of loosely coupled, disparate systems. Indeed, a common defining attribute of a SoS that critically differentiates it from a single monolithic system is the inability to synchronize to achieve common goals, typically due to a lack of interoperability, among the constituent systems. Application of the SoS concept to health care recognizes that it includes myriad stakeholders involving multiple large scale concurrent and complex systems that are themselves comprised of complex systems to address the key challenges as cost effectiveness, functionality and secure, ease of use, [17] advocate a net-centric approach to linking such loosely coupled systems emulating the approach adopted by the Department of Defense. This approach involves linking up hospitals and physicians through networked information systems. Progress is being made along such lines. For example, in Maryland, the Chesapeake Regional Information System Project known as the CRISP project (www.crisphealth.org) is the designated statewide Regional Extension Center and Health Information Exchange helping nearly 1500 providers adopt electronic health care records, securely share health data, and achieve meaningful use. However, net-centricity is not sufficient in itself to achieve overall system goals such as cost reduction while maintaining and improving care. Mittal [2] present an approach to modeling and simulation that directly applies to the coordination structures that must be put into effect to achieve goals at the system of systems level of analysis. Such coordination relies on the ability to interoperate the constituent disparate systems, enabling them to exchange information so that it understood and acted upon by each system to achieve mutual objectives [18]. Systems theory, especially as formulated by Wymore [19], provides a conceptual basis for formulating the interoperability problem of SoS. Systems are viewed as components to be coupled together to form a higher level system, the SoS. The Discrete Event Systems Specification (DEVS) formalism [20], based on Systems theory, provides a framework and a set of modeling and simulation tools [21] to support Systems concepts in application to SoS. 2.3. Personal Limitation Models While systems-based modeling and simulation provides a suitable platform for addressing the SoS problems in health care, there must also be an ability to include human behavioral modeling in order to deal with the limitations that coordinated health care must address. Such agent-based modeling can be viewed as included within agent-directed simulation for systems engineering of large and complex systems. Levent and Oren [22, 23] provide a survey of agent concepts, including agent architectures and taxonomies. From a software development and simulation modeling perspective, agents are advanced objects with intelligent capabilities such as autonomy, and proactive and reactive decision making. A comprehensive account of the system engineering of robot societies and other multiagent systems is given in [24]. However, with the exception of a few studies such as [25], agent-based models have focused largely on replicating intelligent features of human behavior rather than those characteristic of human behavior limitations. Therefore the challenge we address in the proposed methodology is to introduce agent-based models of human behavior limitations needed for coordinated care applications into the SoS level of analysis supported by DEVS-based simulation tools. 2.4. Electronic Health Records Other than user activity models, development of

individualized patient models has been relatively sparse [42], a situation likely to change drastically with the advent of wide spread implementation of Electronic Health Records (EHR) and the push for meaningful use of Health Information Technology [43]. Currently the most widely adopted Health IT message standard for EHRs is the HL7 Version [44]. Unfortunately, as with other standards being considered for meaning use criteria, HL7 v2 messages have no explicit information model, and unclear definitions for many data fields. Although the newer version, HL7 V 3, was developed based on an object-oriented data model it has yet to enjoy widespread adoption. Message Types define the structure and the meaning of each message element (semantics) of the information transfer for a specific messaging. Trigger Events are the explicit sets of conditions that initiate the transfer of information between application roles. Currently, HL7 does not include elements that relate to coordinating care services under consideration here. Thus, as indicated earlier, a criterion for development of the proposed model is to suggest enhancements to electronic health records that become integral to a person s health history. 3. FAMILY OF MODELS APPROACH Based on the above review, the criteria for development of a SoS model for coordinated health care are that it should be: Flexible to meet the variety of stakeholders interests and variety of ACO potential implementations. Scalable to accommodate increases in scope, resolution and detail. Able to integrate system of system concepts system, components, and agent concepts extended to human behavioral limitations. Suggestive of enhancements to Electronic Health Records as needed to support coordinated care. Able to evaluate coordination services and architectures, e.g., patient tracking, medication reconciliation, etc. Such criteria are embodied in the concept of modeling and simulation of SoS [21]. They suggest rather than a single comprehensive model as a feasible solution that a collection of models at different levels of system organization is more practical. Consequently, what is needed is a modeling framework within which to develop such models to form an integrated whole. Recognizing the challenges posed by these criteria we present a disciplined approach to model development that is based on advances in theory and methodology of modeling and simulation over several decades [e.g., 4, 20]. Accordingly, we will discuss developing a framework, based on existing theory and methodology that will guide constructing such a model family. Development of the Modeling Framework takes the following steps: 1. Define the questions to be addressed by the family of models. These questions formulate the objectives of the model development that subsequent work is intended to accomplish. 2. Formalize these questions in terms of a concept of experimental frame which states the conditions under which a model will be experimented with to generate the data intended to address the associated questions. [20]. Experimental frames allow precisely stating a set of requirements for model development, validation, and simulation experimentation. 3. Conceptualize a family of models that can collectively meet the requirements of Step 2 and consequently address the questions in Step 1. The concept of model family recognizes that different questions may be best addressed with different models. 4. The family of models that results from the different levels of abstractions and requirements for efficiency of expression may require different types of models, or modeling formalisms. Recognizing this heterogeneity may require multiple simulation modeling formalisms to co-exist and be interoperated to work together. Such formalisms include differential and algebraic equations (Forrester s System Dynamics, Matlab, Modelica) Finite State Automata (UML Statecharts), and agent based models (MASON, Net Logo), among many others. The theory of modeling and simulation [20], with DEVS as its computational basis is very well situated to provide a foundation for such a methodology. Aumann [4] has shown how such concepts as model specification hierarchy, hierarchical coupled models, experimental frames and morphisms provide a practical methodology for developing credible models of complex systems. Duboz et al. [45] elaborated this methodology emphasizing the roles of experimental frames and model specifications. Recent initiatives such as the RECORD project illustrate the effectiveness of a DEVS based approach [46] in the domain of agro-ecosystem, an SoS domain which shares many of the characteristics of this paper s application domain such as component model coupling and integration of heterogeneous formalisms. Further, the Dynamic Structure DEVS extension [47] allows the structure of the model to change while running which is a fundamental property of SoS. 3.1. Model decomposition and hierarchical design

Aumann [4] defines the nature of the hierarchy of levels of scale based on emergent properties. Based on this concept, he describes the process of specifying a model design in terms that involve assigning entities or units used in the synthesis to the levels in the scalar hierarchical framework. The following is quoted from Aumann s paper [4]. In outline, the process is: 1. Develop a representation of the focal or primary system level by proposing its spatio-temporal limits, describing its main phenomenological characteristics, and specifying the experimental frames operating at this level. In Aumann s example of crab behavior in water low in oxygen and containing clams as well, the focal level was chosen to be that of the individual. Individual crabs have the ability to move, respond to their environment, attack other crabs, feed on prey, etc. For both crabs and the environmental variables, the focal level is a small region of space over which these variables are assumed homogeneous. This region of space keeps track of the age/size structure of crabs and their abundance, and the dissolved oxygen, temperature, depth and salinity at that location. 2. Develop a representation of the sub-systems at the immediately lower level (focal 1) of integration making up the focal level and specifying the relevant experimental frames. In the example, an individual crab consists of submodels governing movement, reproduction, energy balance, aggression, etc., while the way in which the environmental variables change over the small region of space is governed by the environmental inputs and equations governing dissolved oxygen. 3. Develop a representation of the system from the immediately higher level (focal + 1) of integration which the focal level is part of and specifying its relevant experimental frames. In [the] example, we refer to this higher level as the estuary level as it is comprised of all individual crabs, the clams and the environmental variables over the entire estuary. It is at this level that one can talk about population averages for crabs and clams, and also environmental averages like the percent of the estuary low in oxygen over the summer. 4. Defining the processes and relationships linking the focal level to the focal + 1 and focal 1 levels and determining which processes operating between the focal 1 and focal levels are most dependent on the processes occurring between the focal + 1 and focal levels. Information needs to be passed down, up and within each of these levels. For example, as individual crabs move throughout the estuary, the environmental conditions encountered need to be passed to the crab s lower-level models. Alternatively, the location of individual crabs needs to be passed up to the estuary level so that local crab-crab interactions can occur. Health Care Environment Government Policy Business, Technology, Pharmaceutical Trends Coordination Architectures for Accountable Care Organizations Coordination of Individual Care HIT Assisted Coordination of Community Partners Patient Adherence to Care Plan Encourage, Support, Monitor Figure 1. Levels of the Modeling Framework for Coor dination of Health Care Aumann simplifies the formulation of a hierarchical design in terms of a linearly ordered set of three levels. However, this approach can be extended to multiple levels and can recognize that the questions and associated experimental frames need not satisfy a simple linear order (48]. Figure 1 presents a stratification of health care focused on coordination showing four levels. At level 2 we focus on how an individual patient interacts with a specific group of providers for example obstetrics. The immediately lower level 1, allows us to model the patient to predict his/her response to coordination interventions. The next level up expands the focus to populations of patients and groups of provides as formulated in the ACO concept. Continuing the expansion of scope, we recognize the next level up where the environment in which health care operates is not static and needs to be brought in explicitly. The accompanying table (Table 1) outlines the primary stakeholders at each level of Figure 1 and the kinds of questions that they would ask. These questions are operationalized as experimental frames to form the basis for a flexible environment to support multiple stakeholders in health care.

Level Focal Questions Stakeholders Processes and Variables 1. Patient Adherence to Provider s Care Plan 2. Coordination of Individual s Care in a Provider Group (physician, pharmacist, hospital, etc.) 3. Coordinated Care Architectures 4. Health Care Environment How does patient behave in absence of intervention? How does patient react to care guidance? Assuming provider s treatments are 100% effective, what is the effectiveness of coordinating patient s interactions with a group of providers? What is the effectiveness of Health IT-assisted coordination of community health care providers for a given patient population? What are the effects of Environment Change due to Government Policy Business, Technology, Pharmaceutical Trends? Care assistance navigators Strategy designers Architecture Designers/ Accountable Care Organizations/ Medicare negotiators Policy Decision makers, Insurance risk adjusters Patient behavior characteristics, Decision processes Coordination strategies and implementation processes; Quality of treatment outcomes and costs/savings Exploitation of HIT infrastructure networks, medical record standards, information exchange services Table 1. Levels of Health Care Coordination with associated questions, stakeholders and processes 3.2. Model Construction We now discuss applying the above methodology to develop a simulation model at level 3 (Coordinated Care Architectures.) The example is based on the basic framework of Pathways Community Care Coordination developed by the Rockville Institute [49]. In this framework, each pathway is designed to address a single health or social issue and confirm that the issue has been resolved. The Community Care Coordination Hub coordinates care for individuals within targeted medical pathways, such as medication assessment, smoking cessation and obstetrical and postpartum care. The components of the model include the patient s home, the community care coordination hub, and the community partners such as hospital, physicians, pharmacists, etc. The coordination hub is the key addition to the current situation represented by the patient home and community partners. It is intended to fill the vacuum left after the patient has received the formal, though fragmented, care offered by the community partners. The Pathways framework affords the addition of coordination of fragmented services, tracking, monitoring, and assuring that the patient follows physician s directions and medication regimes. The system-level model starts with a top down decomposition (Figure 2) in which a model of the coordinated care system is coupled with an Evaluation Frame component that generates external events for the system and evaluates its responses to them. One of the main objectives implemented by the Evaluation Frame is Figure 2. Top Down Model Decomposition and System Entity Structure of Evaluation Frame to assess proposed pay-for-performance schemes that provide financial incentives that are tied to improving outcomes. The model follows a patient starting with his/her physician through interaction with an array of services provided by the coordination hub. Such hub services include: risk identification, patient tracking, appointment scheduling, transitional care management, medication therapy management, medical care monitoring, home care, and personal health information management. The model will characterize these services to a level of abstraction that creates a space of alternative architectures for the coordination hub. Such architecture alternatives will be assessed in the evaluation frame by

measuring a) a patients health status as they transition through stages of care, and b) accumulating the cost of care (tests, medications, human care managers and providers) including cost of readmissions to hospitals and emergency departments. We formulate architectural alternatives to include those that have been identified as best practice representatives [15]. For example, the careoregon integrated team includes: customer-selected doctor, one or two medical assistants, behaviorist, fulltime nurse providing care coordination administrative assistant providing case management support, specialists and ancillary providers. Such a team is formulated in terms of.a relevant subset of services provided by the coordination hub with characterization of their effect on patient health status and level of resources provided, The coordination services are represented in the model particularly accounting for how health information exchange technology can provide the basis for tracking patients use of hospital and other facilities, keeping track of their compliance with medication regimens and general adherence to the plan of treatment developed by the physician. The model includes a component characterizing the population served by coordinated care. Populations served include homeless, substance abuse, Medicaid at risk, indigenous, mental health, and patients with multiple problems. Figure 3. System Entity Structure for Coordinated Care System The framework employs the System Entity Structure (SES) [21] to enable selection of population served and selection of services provided by the hub to generate appropriate simulation runs to evaluate services in the context of patient needs. This allows evaluation and discovery of architectures that better matches the needs of different populations. The model includes measures of quality of service and cost indicators such as those proposed by Craig et al. [15]. For measuring individual experience of care, the model can include survey instruments such as the site-specific Experience of Care survey, How s Your Health survey items, and Ambulatory Care-Sensitive Hospitalization (ACSH) survey. For the health of a population, survey instruments are available such as Behavioral Risk Factor Surveillance System (BRFSS) and the 12-Item Short Form Health Survey (SF-12). The extensive set of metrics offered by the Healthcare Effectiveness Data and Information Set (HEDIS) will be examined for selective inclusion in the model. Including such an instrument has a concomitant requirement of including corresponding items in patient health information profiles which will be updated with the patients interaction with coordinated care and community partner components. Measures of access to, and utilization of, coordinated care services and community partner resources will arise naturally in the model as it keeps track of patients consumption of these services and use of these resources. Such utilization variables will then be aggregated into to cost outputs. For example, Craig [15] suggests measuring per capita cost by aggregating such quantities as: number of emergency department visits, readmissions, inpatient days, and behavioral health admissions, as well as hospital-based costs (emergency, inpatient, and detoxification).

3.3. Level Specification Given our stated objectives of supporting evaluation of coordinated care of interest to ACOs, it is natural that level 3 (Coordinated Care Architectures) is the focal level. As in Table 1, this allows us to ask questions such as, What is the effectiveness of Health IT-assisted coordination of community health care providers for a given patient population? For example, the family of models should allow architecture designers to explore how exploitation of Health IT infrastructure can enable different coordination architectures With Level 3 as focal level, we descend down to level 2 (Coordination of Individual s Care in a Provider Group) to supply the mechanics needed to produce the desired behaviors at level 2. In this case, we need to replicate experiments, each centered on an individual patient interacting with a subset of community health care partners. The relationship between level 3 and level 2 is that of sampling of patient encounters generated at level 2 from distributions determined by level 3. Generating such a sample encounter takes the following steps in outline: 1. Select a patient from a specified population using SES pruning 2. Select a group of providers to interact with the patient (similar pruning) 3. Simulate a trajectory of the patient within a health care scenario with the provider group (e.g., a nine month pregnancy) 4. Gather statistics on Quality and Cost of care in this encounter 5. Repeat these steps over a set of patients and providers sampled from some specific environment To supply the mechanics of patient behavior in Level 2 we descend to Level 1 (Patient Adherence to Provider s Care Plan) to model the patient as an agent with behavior limitations. To supply the context for Level 3 we ascend to Level 4 (Health Care Environment) where the environmental parameters reside and can be specified. These parameters are employed in lower levels. For example, a poor population in a rural environment determines many of the characteristics (i.e., parameter values) of the patients to be sampled as well as the nature of the providers and services offered. 3.4. Patient Modeling Our approach applies systems concepts (such as state transitions, inputs/output behavior, and coupling of components) uniformly to all components, regardless of their relation to human or non-human elements. A start on such integration is to bring discrete event simulation concepts into multi-agent systems [26]. However, little work has been done on individual, as opposed to population, models [27-28]. An area where the need for models of individual human limitations has been identified is Ambient Assisted Living (AAL) which encompasses technical systems to support elderly people in their daily routine to allow an independent and safe lifestyle as long as possible [29-31]. Our approach is to model the assisted person s particular limitations to support adaptive intervention [32-36]. We model non-adherence with prescription medication due to cognitive and educational limitations [37-41]. 3.5. Dynamic Structure Besides being needed to effectively implement structural changes reflecting reality (e.g., development and birth of a baby), dynamic structure is used to reduce complexity of model structure at different levels. At Level 2 this capability is used to add and remove providers so that only those currently interacting with the patient are included in the active set. Within the patient component, at Level 1, discrete behaviors are rolled in and out so that only those that are currently active are present in the model. For example, behaviors related to interacting with the doctor are present when the patient is coupled to the doctor but not present when interacting with the pharmacy. For large and complex models such as under discussion, such restriction of the field of view helps to increase model comprehension by enabling active components to be the focus of attention while inactive ones are removed from execution. The beneficial reduction in execution time has been discussed in the literature [47]. 3.6. Model Test Bed, Data Calibration and Validation To focus model development and to enable data access for calibration and validation, we are teaming with the Center for Pathways Community Care Coordination (CPCCC), a care coordination nonprofit. The CPCCC is a new program within the Rockville Institute and was launched through a collaborative partnership between the Institute and the Community Health Access Project (CHAP.) CHAP successfully demonstrated how the Pathways Community HUB Model worked with high-risk pregnant women. CHAP will provide a concrete instance of the Pathways Community Care Coordination framework that provides the basis for our SoS simulation model. In the initial phase, our objective is to validate the model when applied to this instance. This is a closed system that is a microcosm representing ambulatory care coordination that enables scaling up to the general case of ambulatory care and further to general truly-integrated health care system as envisioned by the workshop. 3.7. Implementation To implement the methodology discussed above, we are using the MS4 Modeling Environment (MS4 Me ) (www.ms4systems.com), under development by RTSync Corp. The Beta version of MS4 Me with easy-to-use graphic interfaces will be released in public this year, 2012.

The environment is based on the DEVS M&S framework and provides a model-based SoS development facility backed by a full simulation capability. The MS4 Modeling Environment is discussed in detail in [21] 4. CONCLUSION This paper developed a general modeling framework for national health care that will enable simulations to study specific approaches to improve care while reducing cost. The multilevel modeling methodology that was discussed shows how to integrate mathematical system and agent modeling concepts to work across multiple levels from the overall health care environment to patient behavior. The overall model will significantly extend an established coordinated care framework and will be validated against data obtained from a successful program for women with high risk pregnancies. This paper also illustrated how the DEVS and SES formalisms support the modeling methodology by supporting flexible and accurate representation of families of models. Further, dynamic structure was shown to reduce complexity of model comprehension by tracking active components to remain at the focus of attention while inactive ones are removed from execution. The MS4 Modeling Environment supports these capabilities with easy-to-use user interfaces. Beyond impacting health care, the new methodology for SoS modeling and simulation will support constructing and testing alternative architectures for coordinating component systems prior to implementation in diverse societal sectors such as engineering, defense, and agriculture. References 1. Romonko-Slack, L.; Maxwell, C. J; Drugs & Aging 23(4): 345-356; 2006. 2. Instruments Assessing Capacity to Manage Medications ; Farris, Karen B.; Phillips, Beth B.; The Annals of Pharmacotherapy 42(7): 1026-1036; July, 2008. 3. National Health Interview Survey on Disability, Phase 1 and Phase 2 ; National Center for Health Statistics; 1994, 1995. 4. Barea, R., Bergasa, L.M., Lopez, E., Ocana, M., Schleicher, D. and Leon, A., "Robotic assistants for health care," IEEE International Conference on Robotics and Biomimetics (ROBIO 2008), vol., no., pp.1099-1104, 22-25 Feb. 2009. 5. Craig C, Eby D, Whittington J. Care Coordination Model: Better Care at Lower Cost for People with MultipleHealth and Social Needs. IHI Innovation Series white paper. Cambridge, MA: IHI, 2011. 6. Averill Law, W. David Kelton. Simulation Modeling and Analysis, McGraw-Hill Science/Engineering/Math; 3 edition (December 30, 1999 7. N. Wickragemansighe et. al,. Health Care System of Systems; in: Systems of Systems -- Innovations for the 21st Century, Edited by Mo Jamshidi, Wiley, 2008, pp. 542-550 8. B.P. Zeigler, and P. Hammonds, Modeling & Simulation-Based Data Engineering: Introducing Pragmatics into Ontologies for Net-Centric Information Exchange, New York, NY: Academic Press, 2007 9. Tuncer I. Ören and B. P. Zeigler, System Theoretic Foundations of Modeling and Simulation:A Historic Perspective and the Legacy of A. Wayne Wymore, submitted to SIMULATION/ 10. B.P. Zeigler, T.G. Kim and H. Praehofer, Theory of Modeling and Simulation: Integrating Discrete Event and Continuous Complex Dynamic Systems, second edition with co-authors T Academic Press, Boston, 2000. 510 pages 11. B.P. Zeigler and H. Sarjoughian, Guide to Modeling and Simulation of Systems of Systems, Springer, 2012. 12. Andreas Tolk and Adelinde M. Uhrmacher, Agents: Agenthood, Agent Architectures, and Agent Taxonomies in Agent-directed Simulation and Systems Engineering (Levent Yilmaz and T. Oren Wiley Series in Systems Engineering and Management, - VCH; 1 edition (November 17, 2009) 13. Levent Yilmaz and Tuncer I. Ören. Agent-directed Simulation in Agent-directed Simulation and Systems Engineering (Levent Yilmaz and T. Oren Wiley AHRQ and NSF Industrial and Systems Engineering and Healthcare: Critical Areas Workshop, September 21-22, 2009 workshop was co-sponsored by the Agency for Healthcare Research and Quality (http://healthit.ahrq.gov/portal/server.pt?open=512&obj ID=654&PageID=11997&mode=2&sp=true&pubURL =http://wcipubcontent/publish/communities/a_e/ahrq_funded_proj ects/projects/pages/industrial_and_systems_engineering _and_health_care/materials.html 14. Saurabh Mittal, Bernard P. Zeigler, Jose L. Risco Martin, Ferat Sahin and Mo Jamshidi Modeling and Simulation for Systems of Systems Engineering, Systems of Systems -- Innovations for the 21st Century, Edited by Mo Jamshidi, Wiley, 2008, pp. 101-149 15. Sage, A. From Engineering a System to Engineering an Integrated System Family, From Systems Engineering to System of Systems Engineering 2007 IEEE International Conference on System of Systems Engineering (SoSE), 2007 16. G.A. Aumann. A methodology for developing simulation models of complex systems. Ecological Modelling, 202 :385 396, April 2007.

17. http://www.camdenhealth.org/wpcontent/uploads/2011/11/medical-hotspottingcriminals.pdf 18. Gawande A. The hot spotters: Can we lower medical costs by giving the neediest patients bettercare? The New Yorker. January 24, 2011. 19. Older Americans 2010: Key Indicators of Well- Being ; Federal Interagency Forum on Aging-Related Statistics; January 14, 2011. 20. Conwell LJ, Cohen JW. Characteristics of Persons with High Medical Expenditures in the US Civilian Noninstitutionalized Population, 2002. Statistical Brief #73. Rockville, MD: Agency for Healthcare Research and Quality; March 2005. Available at: http://www.meps.ahrq.gov/mepsweb/data_files/publicat ions/st73/stat73.pdf. 21. Olin GL, Rhoades JA. The Five Most Costly Medical Conditions, 1997 and 2002: Estimates for the US Civilian Noninstitutionalized Population. Statistical Brief #80. Rockville, MD: Agency for Healthcare Research and Quality; May 2005. Available at: http://www.meps.ahrq.gov/mepswebdata_files/publicati ons/st80/stat80.pdf. 22. A Randomized Controlled Trial of Fall Prevention Strategies in Old Peoples Homes ; McMurdo, Marion E. T.; Millar, Angela M.; Daly, Fergus; Gerontology 46:83-87; 2000. 23. Medication Nonadherence and Subsequent Risk of Hospitalisation and Mortality among Older Adults ; Vik, S. A.; Hogan, D. B.; Patten, S. B.; Johnson, J. A.; Series in Systems Engineering and Management, - VCH; 1 edition (November 17, 2009) 24. D/Gianni, Bringing Discrete Event Simulation Concepts into Multi-agent Systems pp 186-191,. 10 th Int. Conf. on M&S, 2008 25. Institute of Medicine. Learning healthcare system concepts v. 2008/Annual report. Washington, DC: National Academies Press, 2008 http://www.iom.edu/~/media/files/activity%20files/q uality/vsrt/learning%20healthcare%20system%20c oncepts%20v2008.ashx 26. D/Gianni, Bringing Discrete Event Simulation Concepts into Multi-agent Systems pp 186-191,. 10 th Int. Conf. on M&S, 2008 27. Institute of Medicine. Learning healthcare system concepts v. 2008/Annual report. Washington, DC: National Academies Press, 2008 http://www.iom.edu/~/media/files/activity%20files/q uality/vsrt/learning%20healthcare%20system%20c oncepts%20v2008.ashx 28. Lewis G. Case Study: Virtual Wards at Croydon Primary Care Trust. Available at: http://www.dhcarenetworks.org.uk/_library/croydon_vi rtual_wards_case_study. 29. T. Kleinberger, M. Becker, E. Ras, A. Holzinger, P. Müller, Ambient Intelligence in Assisted Living: Enable Elderly People to Handle Future Interfaces, in Universal Access to Ambient Interaction, vol. 4555, Springer Heidelberg, 2007, pp. 103 112. 30. Kurschl, W. ; Mitsch, S. ; Schoenboeck, J. ; Modeling Situation-Aware Ambient Assisted Living Systems for Eldercare, Information Technology: New Generations, 2009. ITNG '09. Sixth International Conference on Issue Date : 27-29 April 2009 On page(s): 1214 1219 31. Neßelrath, R. ; Haupert, J. ; Frey, J. ; Brandherm, B. ; Supporting Persons with Special Needs in their Daily Life in Smart Home, German Res. Center for Artificial Intell., Saarbrucken, Germany Intelligent Environments (IE), 2011 7th International Conference on, Issue Date: 25-28 July 2011, On page(s): 370-373 32. Dohr, A. ; Modre-Opsrian, R. ; Drobics, M. ; Hayn, D. ; Schreier, G. ; The Internet of Things for Ambient Assisted Living, Information Technology: New Generations (ITNG), 2010 Seventh International Conference on, Issue Date : 12-14 April 2010, On page(s): 804-809 33. Martin, H. ; Bernardos, A.M. ; Bergesio, L. ; Tarrio, P. ; Analysis of key aspects to manage wireless sensor networks in ambient assisted living environments, Data Applied Sciences in Biomedical and Communication Technologies, 2009. ISABEL 2009. 2nd International Symposium on, Issue Date : 24-27 Nov. 2009 34. Hossain, S. ; Valente, P. ; Hallenborg, K. ; Demazeau, Y. ; User modeling for activity recognition and support in Ambient Assisted Living, Information Systems and Technologies (CISTI), 2011 6th Iberian Conference on, Issue Date : 15-18 June 2011, On page(s): 1 4, Print ISBN: 978-1-4577-1487-0 35. World Health Organization, International Classification of Functioning, Disability, and Health (ICF), available at: http://www.who.int/classifications/icf/training/icfbegin nersguide.pdf, 2011, [accessed on February, 2011]. 36. S. Katz, A.B. Ford, R.W. Moskowitz, B.A. Jackson, M.W. Jaffe, Studies of Illness in the Aged. The Index of ADL: A Standardized Measure of Biological and Psychological Function, J. Am.Medical Assoc., vol. 185, 1963, pp. 914 919. 37. J. Frey et al, Towards pluggable user interfaces for people with cognitive disabilities, in Proc. of the 3rd Int l Conf. on Health Informatics. HEALTHINF-2010. Springer, 2010. 38. Brand, M.; Kettnaker, V.; Discovery and Segmentation of Activities in Video ; IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no. 8, p.844-851, 2000.

39. Hongeng, S.; Nevatia, R.; Bremond, F.; Video-Based Event Recognition: Activity Representation and Probabilistic Recognition Methods ; Computer Vision and Image Understanding, vol. 96, no. 2, p.129-162, 2004. 40. Robertson, N.; Reid, I.; A General Method for Human Activity Recognition in Video ; Computer Vision and Image Understanding, vol. 104, no.2-3, p.232-248, 2006. 41. Albanese, M.; Chellapa, R.; Moscato, V.; Picariello, A.; Subrahmanian, V.S.; Turaga, P.; Udrea, O.; A Constrained Probabilistic Petri Net Framework for Human Activity Detection in Video ; IEEE Transactions on Multimedia, vol. 10, no. 6, p.982-996, 2008. 42. William E. Evans, Mary Relling, et al. "Conventional Compared With Individualized Chemotherapy for Childhood Acute Lymphoblastic Leukemia", The New England Journal of Medicine Volume 338 Number 8, 1998 43. HIT Standards Committee, http://healthit.hhs.gov/portal/server.pt?open=512&obji D=1149& parentname=communitypage&parentid=31&mode=2& in_hi_userid=10741&cached=true (accessed Dec 2011). 44. HL7, http://en.wikipedia.org/wiki/health_level_7 (accessed Dec 2011). 45. Duboz R., Bonté B., Quesnel G. 2012. Vers une spécification des modèles de simulation de systèmes complexes. Studia Informatica Universalis. 10: 7-37 46. J.-E. Bergez, P. Chabrier, C. Gary, M.H. Jeuffroy, D. Makowski, G. Quesnel, E. Ramat, H. Raynal, N. Rousse, D. Wallach, P. Debaeke, P. Durand, M. Duru, J. Dury, P. Faverdin, C. Gascuel-Odoux, F. Garcia. An open platform to build, evaluate and simulate integrated models of farming and agroecosystems. Environmental Modelling & Software, in press, 2012 47. F. J. Barros. Modeling Formalisms for Dynamic Structure Systems. ACM Transactions on Modeling and Computer Simulation (TOMACS), 7 :501 515, october 1997. 48. Zeigler BP., James Nutaro, Chungman Seo, Steven Hall, Pamela Clark, Michael Rilee, Sidney Bailin, Thomas Speller, Walter Powell (2012), Frontier Modeling Support Environment: Flexibility to Adapt to Diverse Stakeholders, Symposium on Theory of Modeling & Simulation - DEVS Integrative M&S Symposium, SpringSim, Orlando. 49. www.rockvilleinstitute.org(accessed Aug 2012). Biography Bernard P. Zeigler, Ph.D. is Chief Scientist with RTSync Corp, Research Professor with the C4I Center of George Mason University, and Emeritus Professor of Electrical and Computer Engineering at the University of Arizona. He is internationally known for his 1976 foundational text Theory of Modeling and Simulation, revised for a second edition (Academic Press, 2000), Ernest Lorenza Carter, Jr., M.D., Ph.D. is a senior physician informaticist at Westat with more than 25 years of experience managing and delivering pediatric patient care. He has more than 15 years of experience in the field of health IT. Dr. Carter created and brought to market an Internet-based chronic disease telehealth self-management portal, in addition to overseeing business operations in the area of telemedicine.. At Howard University Hospital, Dr. Carter developed, implemented, and managed telehealth chronic disease management projects in four cities. As director of telehealth science and the Advanced Technology Center at Howard, he created, implemented, and managed projects in telehealth, distance learning, and health IT. Chungman Seo, Ph.D. is a senior research engineer in RTSync Company and a member of Arizona Center for Integrative Modeling & Simulation (ACIMS). He received his Ph.D. in Electrical and Computer Engineering from The University of Arizona in 2009. His research includes DEVS based web service integration, DEVS/SOA based distribution DEVS simulation, and DEVS simulator interoperability. Cynthia K Russell, MSN, RN has over 30 years of nursing experience and is currently a nurse informatics specialist with Westat working with the technical experts assembled on behalf of AHRQ to develop a model EHR Format for Children. Prior to coming to Westat Ms. Russell served as the Emergency and Critical Care Service director for a community hospital. Ms. Russell has participated in a number of health care projects over the years such as establishing an air evacuation service, served as a mentor for the Institute of Healthcare Improvement for medication reconciliation; established a hospital based Rapid Response Team and lead the implementation of an electronic health record for a hospital base prenatal clinic. Brenda A. Leath, MHSA, PMP has a career spanning over 20 years in health services delivery, policy and research. She is a Senior Study Director at Westat, leading translational, dissemination and implementation science research projects. Concurrently, she is the Co-Executive Director of the Rockville Institute Center for Pathways Community Care Coordination where she provides leadership on the development of translation tools, products and metrics aimed at improving service delivery and reducing health disparities. Ms. Leath has held various national appointments, received gubernatorial and mayorial citations, and has been honored federally for her work on health issues of national significance.