Click for next page ( 64


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 63
2 Engaging Complex Systems Through Engineering Concepts INTRODUCTION Along with the increasing interest and concern for the problems sur- rounding health care in the United States has come an increasing aware- ness of the implications of the healthcare system’s complexity. In seeking to engage engineering sciences for insight and strategies for healthcare improvement, it was important to frame the workshop presentations and discussions with a common foundation in and understanding of engineer- ing concepts. The engineering disciplines presented as possible opportunity areas for improving healthcare delivery and management included systems engineering, industrial engineering, operations research, human factors engineering, financial engineering, and risk analysis. William B. Rouse, executive director of The Tennenbaum Institute at Georgia Institute of Technology, described the fundamental perspectives by which systems engineering approaches complex problems. With a particular focus on the nature of prediction, control, and design, Rouse presented a model that shed light on the roots of spiraling healthcare costs and then suggested some likely effects of alternative approaches to controlling costs. Offering a list of standard options from the systems engineering toolbox that might be applied to build processes for controlling costs, as well as some new options described in a Commonwealth Fund report (Schoen et al., 2008), Rouse provided practical insights into how engineers might ap- proach a representative set of issues in health care. Richard C. Larson, Mitsui Professor of Engineering Systems and Civil and Environmental Engineering and director of the Center for Engineering 

OCR for page 63
 ENGINEERING A LEARNING HEALTHCARE SYSTEM Systems Fundamentals at the Massachusetts Institute of Technology, intro- duced some principles of operations research (OR), a systems-oriented ap- proach that draws on the principles of the scientific method to help frame, formulate, and solve difficult problems involving people and technology. Larson offered examples of the application of OR to health care, including work that used sophisticated optimization modeling and computational techniques to advance cancer therapeutics. He said that the techniques of OR have much to offer to the reengineering of systems and processes in health care. He further suggested that the applications of OR and engineer- ing systems with the greatest potential to transform health care have not yet been identified and that further attention is needed to determine opportuni- ties for future progress. Discussing the engineering of systems design tools, James M. Tien, distinguished professor and dean of the College of Engineering at the University of Miami, observed that health care is a complex, integrated collection of human-centered activities that is increasingly dependent on information technology and knowledge. In particular, he explained, health care is a service system. By definition a service system combines three es- sential components—people, processes, and products—and Tien suggested that managing services means, in effect, managing an integrated and adap- tive set of people, processes, and products. He outlined an alternative systems management view of services, discussing the increasing complexity of systems; the increasing need for real-time, adaptive decision making within these systems; and the reality that modern systems are becoming increasingly more human centered. One result is that products and services are becoming both more complex and more personalized or customized. Tien suggested that the methodologies he discussed can be applied to help improve basic services in health care. Essential methodologies of systems engineering were also the focus of a paper by Harold W. Sorenson, professor of mechanical and aero- space engineering in the Jacobs School of Engineering at the University of California, San Diego. Sorenson discussed the principles of an “integrated perspective” for managing complex systems. He outlined the questions that typically apply in engineering complex enterprises, and he described typi- cal approaches that a systems engineer might use to articulate the nature of a problem and to design an appropriate architecture to address it. He provided an overview of how systems engineers think about managing com- plexity, developing solutions, and assessing those solutions. For health care, Sorenson suggested, such an approach could allow a rapid enhancement of capabilities, the development of better working relationships among stake- holders, and the identification of new and more effective ways to deliver patient care—with the potential to lead ultimately to significant changes in healthcare culture, practice, and delivery.

OCR for page 63
 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS CAN WE AFFORD TECHNOLOGICAL INNOVATION IN HEALTH CARE? William B. Rouse, Ph.D., The Tennenbaum Institute, Georgia Institute of Technology The enormous cost of U.S. health care is often cited as a key national challenge (CBO, 2008). Health care is consuming an increasingly large portion of the nation’s gross domestic product (GDP). At the same time, there are concerns that the quality of health care in the United States lags behind that of other countries (IOM, 2000, 2001). It is clear that substan- tial improvements in the delivery of healthcare value are needed, and, it is argued, these improvements should be achievable through value-based competition (Porter and Teisberg, 2006). Of course, it should be kept in mind that our healthcare system did not become the way it is overnight (Stevens et al., 2006). A recent report published by the Congressional Budget Office (CBO) attributes 50 percent of the cost growth in health care over the past four decades to technological innovation (CBO, 2008). Science and engineering research has yielded a steady stream of innovations for detection, diagnosis, and treatment, whose use in many cases has grown by 10 to 15 percent per year. Compounding such growth over 40 years results in a very large level of use. In many domains, such as personal electronics or cellular tele- phones, such growth would be seen as an enormous success. However, the third-party payers of most healthcare bills see this growth as a threat to the viability of the healthcare system. This paper approaches this threat as an engineering problem rather than as a problem of medical science. First, it outlines the engineering ap- proach and contrasts that approach with science. It then explores the CBO’s conclusions a bit more deeply. It proposes three models for controlling the costs of health care so that the growth of these costs tracks the growth in GDP, providing insight into the magnitude of the efficiency gains needed to accomplish this goal. The paper concludes with a discussion of possible ways to achieve these gains. Engineering Approach Determination of the best way to control healthcare costs should be approached as an engineering problem rather than as a medical sci- ence problem. The potential of engineering to enhance health care has, of late, received increasing attention (NAE/IOM, 2005). This potential can be understood in terms of the following levels of understanding of any phenomenon:

OCR for page 63
 ENGINEERING A LEARNING HEALTHCARE SYSTEM • Describe past observations. • Classify past observations. • Predict future observations. • Control future observations. • Design future observations. Science progresses from describing and classifying past observations to predicting future observations. If these predictions turn out to be accurate, science concludes that the theory or model employed has credence. If not, the theory or model needs revision. The goal is to create valid knowledge. Engineering builds on scientific knowledge, particularly in using mod- els to predict. However, engineers usually are not content just to predict. They also want to control the state of the system of interest or, if they can, to design or redesign the system to facilitate better control. In some cases, this penchant for design and control has enormous societal implications (McPhee, 1990). Predict Taken simplistically, there are two basic approaches to prediction. One is extrapolation. Equations are fit to data collected under particular condi- tions. These equations are then used to project the outcomes for similar conditions. Statistical models, such as those used in medicine for random- ized controlled trials (RCTs), are examples of equation fitting. To the extent that the conditions of the trials adequately reflect the eventual conditions of use, we can be reasonably confident that similar outcomes will be attained when a treatment moves from trials to clinical use. RCTs work well, although slowly and expensively, when there are large populations that can be observed under controllable conditions. However, this approach cannot be employed for the study of large-scale systems such as health care. There simply are not enough healthcare systems to achieve statistical significance in a study of the large, systemic changes likely to be needed to control costs and enhance quality to the extent outlined earlier. Engineering approaches to solving large-scale problems typically rely on models as a basis for prediction. These models are formulated from “first principles” drawn from a range of scientific domains. These prin- ciples, usually stated as fairly simple mathematical relationships, become elements of much larger mathematical and computational models that are used to predict the outcomes of different approaches to the design and control of complex systems. Engineering approaches are illustrated later in this paper.

OCR for page 63
 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS Control Engineering the control of a system involves measurement, feedback, and compensation to achieve system objectives. Measurement is used to ascertain the state of the system. This, of course, requires defining system state variables, their units of measure, and how such measurements can be made. Feedback involves comparing predicted and actual system states in order to correct errors. Such feedback results in a “control loop.” Com- pensation concerns adding dynamic elements to the control loop in order to counteract delays and lags in system response. Design Engineering design involves problem analysis, solution synthesis, pro- duction of an artifact that embodies the solution, and then sustainment of the system in its use. Analysis involves understanding input–output relation- ships, including uncertainties, and then creating models, as discussed above. Synthesis is a matter of designing input–output relationships to achieve system objectives. Production involves the various actions—fabrication, construction, programming, and so forth—necessary to create systems that embody the desired relationships. Finally, sustainment concerns creating mechanisms that ensure that system objectives will be met in the future. Summary Engineering approaches to prediction, control, and design have much to offer health care with respect to making systemic improvements by de- creasing costs and increasing quality. The remainder of this paper provides an illustration of how engineering might help in meeting the challenges faced by health care. Healthcare Illustration As discussed earlier, the past four decades have seen enormous increases in healthcare costs. Specifically, real healthcare costs tripled as a percentage of GDP in the period from 1965 to 2005, with half of this growth due to technological innovation (CBO, 2008). The magnitude of these increases has led some to conclude that the healthcare system is “running on empty” (Peterson, 2005). There appears to be virtually unanimous agreement that the system must change significantly. Figure 2-1 summarizes the overall phenomenon discussed in the CBO report. Technological inventions become market innovations as they in- crease in effectiveness and the associated risks decrease. The result is in-

OCR for page 63
8 ENGINEERING A LEARNING HEALTHCARE SYSTEM Technology Innovation Decreased Increased Increased Efficiency Risk Effectiveness Longer Decreased Increased Cost/Use Use Life Increased Expenditures Improved Care FIGURE 2-1 The dynamics of escalating healthcare costs. F2-1.eps creased use, which in turn leads to increased expenditures. In parallel, increased efficiency through production learning (discussed further below) leads to decreased cost per use, although not enough to keep pace with the product’s growing use in health care. Finally, increased use yields improved care, which leads to longer lives and increased chances of again employing the technology of interest. The concern in this illustrative example is how to control the phenom- enon depicted in Figure 2-1. In typical engineering fashion, we approach this control problem with a series of models, beginning with a very simple model and then elaborating as the limits of each model become clear. Model 1: Growth The first model considers what efficiencies are needed to counteract the growth in Figure 2-1. We start with a simple equation: Cost (1 – α) Use (1 + β) = Total (1 + δ), (1)

OCR for page 63
9 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS where α is the annual rate of cost reduction, β is the annual rate of usage growth, and δ is the annual allowable total growth. A bit of algebra shows that the annual rate of cost reduction required is given by the following: α = (β – δ)/(β + 1). (2) Table 2-1 shows the cost reductions needed for five of the technologies discussed in the CBO report, assuming zero allowable growth. These are rather significant decreases. However, these decreases are more instructive than definitive because of the simplicity of the model. In particular, the model is quite limited in that it provides no mechanism for achieving cost reductions and does not differentiate between the various elements of the healthcare delivery process. Thus we need to elaborate on model 1. Model 2: Learning The second model considers production learning, a well-understood concept in industrial engineering (Hancock and Bayha, 1992). Quite sim- ply, as one produces more of an item, one gets better at it, and unit costs decrease. In some industries, such as the semiconductor industry, these decreases are a primary source of profit margins. Many manufacturing industries employ production learning curves to predict costs and hence profits. The basic learning equation is given by Cost (t = T) = Cost (t = 0) No. Uses (t = T)-Rate. (3) This learning phenomenon is usually discussed in terms of “percent curves.” For example, a 70 percent curve means that after each doubling of the number of units produced, unit costs drop to 70 percent of what they were after the previous doubling. Table 2-2 provides a few examples of the rates required in equation (3) to achieve different percent curves. TABLE 2-1 Cost Reductions Needed to Accommodate Growth Annual Rate Minimum Annual Rate Treatment of Usage Growth (%) of Cost Reduction (%) Angiography 10 9 Angioplasty 15 13 Dialysis 12 11 Hip replacement 10 9 Knee replacement 11 10

OCR for page 63
0 ENGINEERING A LEARNING HEALTHCARE SYSTEM TABLE 2-2 Production Learning Parameters Percentage Cost Per Use Rate for for Each Doubling of Uses Learning Model 70 0.515 80 0.322 90 0.152 Most learning curves fall in the 70 to 90 percent range. This range re- flects the experiences of many industries, including producers of airplanes, automobiles, and electronics. Curves below 70 percent are rare. As the results given below will show, controlling healthcare costs may require achieving significantly below 70 percent—a significant challenge. Figure 2-2 shows learning curves for the three learning rates in Table 2-2, assuming a 10 percent annual rate of growth in usage. Note that the initial conditions were 100 uses at $100 per use, yielding an initial total expenditure of $10,000. Figure 2-3 shows the growth of total expenditures, again assuming a 10 percent annual growth in usage. Table 2-3 shows the overall results for annual growth rates of 5 and 10 percent, assuming a 70 percent learning curve. Unit costs have dropped significantly, but the growth in usage has overwhelmed these efficiencies. Overall, this model exhibits impressive cost reductions from production learning, but it does not indicate where or how this learning happens. Furthermore, the model does not reflect the process whereby health care is delivered. 10 0 80 Costs Per Use 60 70 % 40 80 % 20 90 % 0 4 4 4 1 6 7 0 3 9 9 6 5 0 92 2 31 81 21 12 14 17 67 25 45 10 55 98 38 4 11 14 Uses FIGURE 2-2 Learning curves for the three learning rates from Table 2-2. F2-2.eps

OCR for page 63
1 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS 50,0 00 40,0 00 Total Costs 30,0 00 20,0 00 10,000 0 11 1 3 5 7 9 17 13 15 19 21 23 25 27 29 30 Year FIGURE 2-3 Expenditure growth at 10 percent annual growth in use. F2-3.eps TABLE 2-3 Impacts of Production Learning Results at 30 Years Rate (%) No. of Uses Cost/Use ($) Total Expenditures ($) 5 412 48 19,874 10 1,586 24 38,256 Model : Process The third model explicitly considers the process by which healthcare service is provided. As shown in Figure 2-4, this process includes multiple stages and differentiates labor from technology. A rich experience base al- lows us to define the learning rates for technology. For present purposes we, somewhat optimistically, set the technology learning rate at 70 percent. The question then is, What labor learning rate is needed to control the growth in costs to an acceptable level? This model is given by the follow- ing equations: (4) Cost (t) = Cost of Labor (t) + Cost of Technology (t), (5) CTOT (t) = CPUL (t) NU (t) + CPUT (t) NU (t), CPUL (t) = CPUL (1) NU (t)-RateL, (6) CPUT (t) = CPUT (1) NU (t)-RateT, and (7) NU (t) = NU(1) (1+β) t-1. (8) where CTOT, CPUL, and CPUT denote total costs, labor cost per unit, and technology cost per unit, respectively, while NU denotes number of units. Figure 2-5 shows the efficiency required to control increases in health-

OCR for page 63
2 ENGINEERING A LEARNING HEALTHCARE SYSTEM Labor Labor Labor Labor Diagnosis Treatment Recovery Detection Technology Technology Technology Technology FIGURE 2-4 Service delivery process model. F2-4.eps 80% 70% Labor Cost Per Use 60% 50% 40% 30% GDP = 0% 20% GDP = 2% GDP = 4% 10% 0% 5% 10% 15% Technology Use Grow th Rate FIGURE 2-5 Required efficiency (% cost per use per doubling) for healthcare costs F2-5.eps to track gross domestic product (GDP). care costs to the point that they track increases in the GDP. The best case is for 4 percent GDP growth and 5 percent usage growth, which requires a learning curve of greater than 70 percent for labor. This magnitude of learning is imaginable. The worst case is for 0 percent GDP growth and 15 percent usage growth, which would require a learning curve of greater than 40 percent. This level of learning has never been achieved in any domain. Implications The implications of the results of these three models are quite clear. To limit the growth in total healthcare spending to the growth in GDP, some combination of the following three things is needed:

OCR for page 63
 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS • Limit the growth of technology use. • Limit the cost of technology use. • Decrease the cost of labor associated with technology use. Overall, the savings due to learning are the key to affordability. Achiev- ing these savings will, however, be a significant challenge since learning rates of less than 70 percent are difficult to achieve. Sources of Learning In industries in which production learning curves have long been used, the sources of learning include labor efficiency, changes in personnel mix, standardization, specialization, method improvements, better use of equip- ment, changes in the resource mix, product and service redesign, and shared best practices. The Commonwealth Fund recently published recommenda- tions for “bending the curve” (Schoen et al., 2008). Based on extensive economic analyses, the following are recommended as ways to reduce healthcare costs: • Producing and using better information − Promoting health information technology − Center for medical effectiveness and healthcare decision making − Patient shared decision making • Promoting health and disease prevention − Public health: reducing tobacco use − Public health: reducing obesity − Positive incentives for health • Aligning incentives with quality and efficiency − Hospital pay-for-performance − Episode-of-care payment − Strengthening of primary care and care coordination − Limit on federal tax exemptions for premium contributions • Correcting price signals in the healthcare market − Resetting of benchmark rates for Medicare advantage plans − Competitive bidding − Negotiated prescription drug prices − All-payer provider payment methods and rates − Limit on payment rate updates in high-cost areas The report Bending the Cure provides projections of the savings that could be realized by adopting these recommendations (Schoen et al., 2008).

OCR for page 63
10 ENGINEERING A LEARNING HEALTHCARE SYSTEM environment. If this methodology is successfully applied, the changes that result can have the effect of altering the culture of healthcare delivery, with concomitant changes in the practice and delivery of care. General Approach When we approach the problem of engineering complex enterprise systems, we typically start by asking a series of interrelated questions. How do we think about the problem? How do we manage the complexity? How do we approach the problem’s solution? How do we develop the problem’s solution? How do we assess the proposed solution? What is the effect of the development? Each of these questions is discussed briefly in the sections that follow, with a particular emphasis on the first three. How Do We Think About the Problem? Any enterprise, whether a healthcare or business or defense enterprise, involves a large number of people with a variety of responsibilities and jobs. To carry out these responsibilities and jobs, these stakeholders work using prescribed processes to accomplish desired functions and outcomes. And in performing this work, they use a variety of information and data. In most enterprises, the reality is that any given user is familiar with only a lim- ited number of processes and data sources. Too often this implies that the function being performed has more limited utility than would be the case if the workers’ knowledge of processes and data were broader and more encompassing. Because of the stovepipe or silo characteristics embedded in virtually every organization, a given user in an enterprise is unfamiliar with all the other useful components of the system and has no way to learn them in a natural way. How do we bring order to such a system? How do we manage this environment in such a way that the users can actually use the data, the processes, and the functions that exist elsewhere in the enterprise? In much of his writing, Russell Ackoff (a seminal thinker in systems science at the Wharton School) emphasizes the need to break away from reductionist thinking in managing complex enterprises. He goes to the heart of the ques- tion in his book On Purposeful Systems, written with Frederick Edmund Emery (Ackoff and Emery, 1972), when he writes, “To manage a system effectively, you might focus on the interactions of the parts rather than their behavior taken separately.” Indeed, the issue is all about interactions, interfaces, and the way information is shared and distributed. A long-time collaborator with Ackoff, Jamshid Gharajedaghi, outlines the basics of systems thinking in his book Systems Thinking: Managing Chaos and Complexity (Gharajedaghi, 1999). Such thinking starts, he

OCR for page 63
10 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS suggests, with an operational definition of a “systems methodology” that involves three interdependent variables: structure, function, and process. Those variables, together with the environment, define the whole. For pres- ent purposes, we can think of structure as defining the components and their relationships and constraints—synonymous with input, means, and effects. Similarly, function defines the outcome, which is synonymous with output, and process defines the sequence of activities required to produce the outcome—how the function is performed. A core assumption is that the development process is necessarily iterative. As a quick example of the interdependency of function, structure, and process, consider the heart. The function of a heart is to pump. It circu- lates blood. Its structure can be defined in terms of its chambers, valves, and arteries. The process is basically defined in terms of alternating cycles of contraction and expansion. The environment is the body in which the heart operates. The body in turn operates in a larger environment, which certainly affects the functioning of the heart. Clearly, the variables are inter- dependent and cannot be separated in considering the heart as a system. Figure 2-9 provides background on the evolution of systems thinking. A product of the competitive world in which we live, systems thinking began with the ideas of Henry Ford regarding mass production. His model posited that people and parts are interchangeable, a way of thinking that Multiminded Mindless System Uniminded System System SHIFT Social Model OF Purposeful Machine Model Biological Model PARADIGM Society Organization Members Interchangeability Diversity and Analytical Participative of Growth Approach Management Parts and Labor Self-organizing Systems Independent Variables Social systems are Henry Ford’s Alfred Sloan’s information-bonded Mass Production Divisional Structure Tavistock Institute’s System Socio-Tech Model Systems Approach Joint Optimization Flexibility and Redesign Control Interdependent Ford’s Whiz Kids Ohno’s Lean Ackoff’s Interactive Variables Production Management Operations Research Cybernetics Model Choice FIGURE 2-9 Evolution of systems thinking. Figure 2-9 redrawn

OCR for page 63
108 ENGINEERING A LEARNING HEALTHCARE SYSTEM led to people and parts being treated as independent variables. The analyti- cal result was to look at each part by itself. As people first experimented with mass production, they learned that the factors they regarded as being independent actually were not, but rather were interdependent. This inter- dependence led to a systems approach, which can be said to be the roots of operations research. The analytical problem became a search for optimal solutions that would enable an organization to do things better. An oft-cited example of the acceptance of the OR approach is the scenario involving the “whiz kids” at Ford Motor Company around 1960. Ford started with the motto that one could have any color Ford one wanted as long as it was black. As mass production gained in popularity, it became apparent that everybody could institute mass production tech- niques; as a result, the first adapters of mass production, such as Ford, would lose their competitive edge. Alfred Sloan started General Motors (GM) with the idea that diversity would lead to growth. Buyers of GM products could get cars with different colors and in different varieties. To manage the resulting interdependencies, Sloan introduced a reductionist approach that was reflected in the divisional structures of the company. Or- ganizational structures were broken down and reduced to the point where the company considered marketing to be separate from human resources, which was separate from sales, and so forth. That led to the divisional structure that is still taught in management schools today. This is much more of a biological model than Ford’s mass production model. It is a biological model because there is a brain (i.e., the chief executive officer) in charge who tells his or her arms (i.e., the divisional structure) what to do. The divisions implemented the strategy from the corporate office to develop diverse products and to stimulate corporate growth and profit. This business model evolved to another stage with the concept of lean production, developed by Ohno at Toyota. Lean production is based on a cybernetics model in which one measures the production process and feeds that information back into decisions about inventory and production to gain efficiencies and reduce costs. This is essentially the state of the practice in most organizational theory today. The past decade, however, thanks in large part to the power of the Web, has seen the rise of participative management. The person in charge no longer has the sole word. Based on personal experience working at high levels in companies, I can say that collaboration and participation provide the essential mechanisms to manage large, complex enterprises effectively. Top-down direction is increasingly complemented by bottom-up involve- ment in the decision-making process. A fundamental characteristic of this new, productive environment is its social nature. The power of these social networks is a primary source of the complex adaptive systems we are start- ing to recognize and address.

OCR for page 63
109 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS Social systems are information bonded, in contrast to the systems of the past, which were energy bonded. Consider the car. It has an engine, and when the car is started, the gas in the engine explodes, and the energy from that explosion is transferred through the camshaft to the driveshaft, to the axle, and to the wheels. Everything is connected by various forms of energy transfer. Most electric power grid systems operate the same way, with the electrical energy being transferred in ways that allow us to do what we want. Energy-bonded systems are created by well-known systems engineer- ing processes that are based on well-developed requirements. In essence, it is the statement of requirements that separates project management from corporate management. In the case of a technology, its usefulness is the primary developmental concern, but enterprise systems must be concerned with organizational interactions and with the people involved in them. Learning healthcare systems can be described as being information bonded and operating through social networks. The social model basically says that if one is purposeful and seeks to reach some kind of enterprise objective, one should take a holistic view that includes society, the organization, and, finally, the various members of that society or organization who may be involved in what one is trying to do. This holistic set of concerns leads to a totally different approach from the traditional systems engineering, which is reductionist in nature. Now, instead of analyzing things, one designs, one tries out, one builds, one fields, and one learns. The development proceeds in an iterative fashion, with new capabilities appearing on a time scale of days or a few weeks. Collabora- tion among stakeholders and managers becomes the necessary development interaction style. Often no single person is in charge in the sense of having tight control over the development plan. Flexibility and adaptation are central to the process and the participative management style. To make an important point, let me introduce a little engineering jar- gon. Entropy is the measure of randomness in the universe. According to the Second Law of Thermodynamics, total entropy in a closed system increases over time. However, open-living systems display an opposite tendency; they move toward order, thus generating negative entropy. Emergent behaviors are a result. The Internet is a good example. Think of all the behaviors that have emerged from the use of the Internet. I am sure the creators of the Internet had no idea of the sorts of behaviors that would arise and garner so much interest, such as the information-sharing power of Google, the online auctions of eBay, and communication sites such as MySpace. These emergent behaviors are the key to understanding the potential for changing the culture, practice, and delivery of health care.

OCR for page 63
110 ENGINEERING A LEARNING HEALTHCARE SYSTEM How Do We Manage the Complexity? Data and information must be at the heart of any discussion of information-bonded systems. Consequently, the methodology being de- scribed here starts with a consideration of the mechanisms for making data and information available across an enterprise in ways that are effectively transparent to the system user. In this approach, sources and generators of data and information publish to a registry that is used to make potential users aware of the existence of the data and information and that has pointers defining how the data are found and retrieved. The users, gener- ally, subscribe to the types of data and information they are interested in accessing. The mechanism for accomplishing this publish/subscribe opera- tion is referred to as a serice registry. The registry is important because it guarantees that data will not be warehoused. Ownership and responsibility for the data remain with those who generate them, and once the registry has been established, the methodology says that information is retrieved, processed, integrated, and managed to achieve a purpose. If new informa- tion is generated, it is registered for future users of the results. The pointers are provided so that subscribers can quickly get the information needed to solve a problem or answer a question. Alternatively, Web standards allow the discovery of information that supplements the subscriber information that is registered. For example, Google provides a search engine people use regularly in their daily lives. Standards that have already been developed for the Web are fundamental to the development of the mechanisms used to define interfaces and govern the exchange of data and information. The processes for dealing with data and information form the basis of the methodology. Enterprises must adapt and respond in a timely and effective manner to a variety of planned and unplanned events. To deal with this ever-present situation, it is important that there be a seamless interoperation of disparate organizational entities, often in unplanned and complex ways. The rapidly changing environment has led to a search for solution inariants that might provide stability for the creation and evolution of the system that supports information interoperation (referred to here as the enterprise knowledge system [EKS]). The current approach for managing enterprise complexity, which is continuing to evolve and mature, is based on three interdependent variables: structure, function, and process. The structure, which we refer to as an architecture, provides a blueprint for evolving the EKS. It defines the components or actions of the system and their functional interconnections. The architecture is also used to capture the constraints on the system (e.g., quality of service, security and privacy rules, regulatory constraints such as HIPAA). The structural description does not involve detail on the imple- mentation of the components, communications, or constraints. Instead, it

OCR for page 63
111 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS should be developed relatively quickly and should provide a documented basis for the implementation and evolution of the EKS itself. As with the design of a building, the architecture provides a reference that changes less often than the implementation and permits a framework within which processes and functions are allowed to change to accomplish the objectives of the business. How Do We Approach the Problem’s Solution? Deriving and defining the architecture is the key step of the develop- ment process for complex, adaptive systems. Fundamental to the design process is the concept of separation of concerns. Generally, the first step in separating concerns is to assume that the architecture is defined as con- sisting of layers or tiers. Each layer communicates with its adjoining layer through well-defined interfaces. An activity in one layer cannot interact directly with any internal feature of any other layer. This structure allows changes to be made within a layer without disturbing other layers as long as the interface descriptions remain unchanged (i.e., concerns are separated). A fundamentally important layer is the presentation layer. Users interact with the system using a portal, sometimes referred to as the human–system interface. For our architecture construct, users, through the portal, have ac- cess to the service registry and the elements of the system they are interested in using. Users can ask questions and, essentially, tell the system what they are trying to do. Everyone should be familiar with the function of a portal through Web interactions. For example, the AOL browser serves as a portal to the global community and its services. Behind the registry is an infrastructure that allows the complexity of the networked system, such as the Web, to be hidden and transparent to the user. Through this common information infrastructure (CII), informa- tion that addresses the needs of a user is routed to appropriate functions, applications, processes, and services. An architectural style that is gaining wide acceptance is serice-oriented architecture (SOA). In this style, the CII is generally referred to as the enterprise serice bus. A very simple model of architecture has users who can access the sys- tem through various kinds of hardware devices and communities of interest that are defined through having a common interest or mission that requires a variety of business applications or services. Underpinning this architecture is the CII. In this simple model of an enterprise architecture, it is important to recognize that a network simplifies the system connections of many us- ers and their applications. A set of N users can communicate directly with one another without being connected in a point-to-point fashion. The lat- ter mode of communication would generate an enormously complicated requirement because the number of point-to-point connections for N users

OCR for page 63
112 ENGINEERING A LEARNING HEALTHCARE SYSTEM is proportional to the square of the number of users. With a networked system, the number of connections is simply N. Each user connects to the network, and the network accomplishes the point-to-point connectivity. Returning to the SOA, the term serice denotes a broad set of useful artifacts that enable users to accomplish their tasks. A service can be an application that may be useful for people across the enterprise, it can be data sources or databases that have widespread utility, or it can be com- putational tools that support implementations invisible to the user. In es- sence, a service is a reusable artifact that simplifies the development and implementation of, and facilitates the introduction of, new processes and functions that broaden the existing system capability. The SOA model is based on the differing needs and roles of service consumers and providers. The consumers interact with the system through a presentation layer. The business processes must be precisely defined and are used to derive essen- tial services. The composition of the services produces the desired process. Required services are located via the service registry, wherever they may be located physically. The registry is used to inform the service provider of the request for the service, and the enterprise service bus executes and allows the execution of the desired business process. In summary, one starts by asking, “What are we going to do in our business?” In the present case we might ask, “How are we going to deliver health care?” One then defines the processes that describe a healthcare func- tion. From these processes and from the actors using the system, services are defined and added to the service registry. As described, the architecture of the system is layered, and an emphasis in the system development is on the interfaces between layers. One can change a business process without changing the service or presentation layer except in the details of that pro- cess. The value of managing the complexity of the EKS through the use of layered, or N-tiered, architectures then starts to make sense. How Do We Develop the Problem’s Solution? Having defined a framework for managing the complexity, it is logical to ask, “How do we develop the problem solution?” First, one identifies a function or mission that is to be provided by the healthcare delivery system. Then one identifies the people, or actors, who should be involved in us- ing, describing, architecting, and implementing the desired capability. This group of people constitutes a community of interest, and the members are referred to as stakeholders. The future users of the system are requested to answer the question, “As an actor, how am I going to use this system?” The answers are called use cases, which become the basis for discussions between the architect and the users as the architecture is developed. The architect guides the discussion but must always be sensitive to the needs of

OCR for page 63
11 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS the user community. For the healthcare delivery community, the use cases should focus on the needs of and the services desired by and delivered to each patient. They are key actors for this system development. The manner in which the architect translates the use case information into the structure of the desired system is not addressed here, but the architec- ture must provide guidance to the engineers who implement the system. The architect implements governance procedures that enable the implementation to be measured against the requirements of the architecture. A key part of the governance process is continuing interaction among the stakeholders, archi- tects, and implementers. This interaction drives the iterative development of the system and must serve to put in place useful capabilities, often in small increments, on a frequent basis. The implementation approach is referred to as agile deelopment. Practice has revealed a desirable phenomenon that often emerges from providing new capabilities in small, easily understood increments. Because the users can live with and use the capability, a fan club develops that facilitates the adoption of the capability across a broader community. This is one example of a desirable emergent behavior that can appear when one is working within a highly complex environment. These behaviors are observed to change cultures in a natural manner that comes from bottom-up participation, not from the top down. How Do We Assess the Proposed Solution? Early in this paper, I mentioned the importance of feedback control and its great utility in virtually all systems and their useful operation. There has been no discussion of the role of feedback control in this methodology, but there was an implicit reference in the previous paragraph. In essence, it is through various types of feedback that we address the question, “How do we assess the proposed solution?” The development of complex systems such as healthcare delivery ad- vances because of feedback and communication among all the participants. The architect serves as the feedback controller during the development of the architecture. Because the systems being developed are inherently complex, it is often difficult for the architect to determine whether the logic and behaviors dictated by the architecture will achieve the goals of the stakeholders. Systems of the type discussed here are event driven. Us- ing theory and methods drawn from discrete-event dynamic systems, the architect can develop models of the capability being developed that can be used to assess the adequacy of the logical architecture and the behavior of the processes implemented by the architecture. These architectural models do not involve the actual implementation of any capabilities. They can be applied before time and resources are expended to actually build the sys- tem. In fact, they are useful in saving later development costs and delays in

OCR for page 63
11 ENGINEERING A LEARNING HEALTHCARE SYSTEM situations where fundamental problems are discovered late in the process. Finally, the architect serves as the controller in the governance policies that are imposed on the implementers. As useful capabilities are presented to the user community, the architect again assesses the reactions of the users to identify desirable modifications, additions, and improvements. As a capability moves into daily use, however, the architect must have planned for the possibility of behaviors arising either from events outside the community of interest or from unplanned events occurring within the system that can adversely affect the system’s usefulness. To anticipate the ef- fects of complex events on the desired system, the architect must introduce mechanisms for measuring the internal message traffic. Sometimes referred to as business activity monitoring, these measurements are used to provide feedback that allows the earlier identification, recognition, and correction of anomalous behaviors. Thus, feedback control becomes essential in the long-term operation and performance of the system. What Is the Effect of the Development? In closing, we consider the question, “What is the effect of the devel- opment?” This paper has asserted that it is possible to build an environ- ment that will lead to the rapid fielding of enhanced and new capabilities. This result can be achieved only through close working relationships among all stakeholders, including patients, healthcare administrators and practitioners, and enterprise architects and engineers. Not only can there be planned developments that provide more effective ways of delivering health care, but unexpected and useful emergent behaviors can appear. As a result, the culture, practice, and delivery of patient care will change fundamen- tally in important and beneficial ways. This change will be driven from the bottom up by the participants while being guided by the leadership of the healthcare community. REFERENCES Ackoff, R., and F. E. Emery. 1972. On purposeful systems. Chicago, IL: Aldine-Atherton. Brandeau, M. L., F. Sainfort, and W. Pierskalla (Eds.). 2004. Operations research and health care. A handbook of methods and applications. New York: Springer. Bush, G. W. 2001 (October 16). Executie order on critical infrastructure protection. Wash- ington, DC: The White House. CBO (Congressional Business Office). 2008. Technological change and the growth of health care spending. Pub. No. 2764. Washington, DC: CBO. Friedman, T. L. 2005. The world is flat: A brief history of the twenty-first century. New York: Farrar, Strauss & Giroux. Gawande, A. 2002. Complications: A surgeon’s notes on an imperfect science. New York: Metropolitan Books.

OCR for page 63
11 ENGAGING COMPLEX SYSTEMS THROUGH ENGINEERING CONCEPTS Gharajedaghi, J. 1999. Systems thinking managing chaos and complexity: A platform for designing business architecture. Boston, MA: Butterworth–Heinemann. Grossman, J. H. 2008. Disruptive innovation in health care: Challenges for engineering. The Bridge 38(1):10–16. Hancock, W. M., and F. H. Bayha. 1992. The learning curve. In G. Salvendy (Ed.), Handbook of industrial engineering, 2nd ed. (pp. 1685-1698). New York: Wiley. Hipel, K. W., M. M. Jamshidi, J. M. Tien, and C. C. White. 2007. The future of systems, man and cybernetics: Application domains and research methods. IEEE Transactions on Systems, Man, and Cybernetics Part C 30(2):213–218. IOM (Institute of Medicine). 2000. To err is human: Building a safer health system. Washing- ton, DC: National Academy Press. _____. 2001. Crossing the quality chasm: A new health system for the 21st century. Washing- ton, DC: National Academy Press. Kaplan, E. H., and R. J. Heimer. 1992. HIV prevalence among intravenous drug users: Model- based estimates from New Haven’s legal needle exchange. Acquired Immune Deficiency Syndrome 5(2):163–169. Kaplan, E. H., D. L. Craft, and L. M. Wein. 2002. Emergency response to a smallpox at- tack: The case for mass vaccination. Proceedings of the National Academy of Sciences 99(16):10935–10940. Maglio, P., S. Srinivasan, J. Kreulen, and J. Spohrer. 2006. Service systems, service scientists, SSME, and innovation. Communications of the Association for Computing Machinery 49(7):81–85. McPhee, J. 1990. The control of nature. New York: Farrar, Straus and Giroux. Mendonca, D., and W. A. Wallace. 2004. Studying organizationally-situated improvisation in response to extreme events. International Journal of Mass Emergencies and Disasters 22(2):5–29. Morse, P. M., and G. E. Kimball. 1951. Methods of Operations Research. Cambridge: Tech- nology Press of Massachusetts Institute of Technology. NAE (National Academy of Engineering)/IOM. 2005. Building a better deliery system: A new engineering/health care partnership. Washington, DC: The National Academies Press. O’Neill, P. 2007. Why the U.S. healthcare system is so sick and what OR can do to cure it. OR/MS Today 34(6). www.lionhrtpub.com/orms/orms-12-07/frhealthcare.html (accessed May 28, 2010). Peterson, P. G. 2005. Running on empty. New York: Picador. Porter, M. E., and E. O. Teisberg. 2006. Redefining health care: Creating alue-based competi- tion on results. Boston, MA: Harvard Business School Press. Quinn, J. B., J. J. Baruch, and P. C. Paquette. 1987. Technology in services. Scientific American 257(6):50–58. Rouse, W. B. 2007. Complex engineered, organizational and natural systems. Wiley Inter- Science Online: Systems Engineering 10(3):260–271. _____. 2008. Health care as a complex adaptive system: Implications for design and manage- ment. The Bridge 38(1):17–25. Schoen, C., S. Guterman, A. Shih, J. Lau, S. Kasimow, A. Gauthier, and K. Davis. 2008. Bend- ing the cure: Options for achieing saings and improing alue in U.S. health spending. New York: The Commonwealth Fund. Stevens, R. A., C. E. Rosenberg, and L. R. Burns. (Eds.) 2006. History and health policy in the United States. New Brunswick, NJ: Rutgers University Press. Sussman, J. M. 2008. Intelligent transportation systems in a real-time, customer-oriented society. The Bridge 38(2):13–19.

OCR for page 63
11 ENGINEERING A LEARNING HEALTHCARE SYSTEM Tien, J. M. 2000. Individual-centered education: An any one, any time, any where approach to engineering education. IEEE Transactions on Systems, Man, and Cybernetics Part C: Special Issue on Systems Engineering Education 30(2):213–218. _____. 2003. Towards a decision informatics paradigm: A real-time, information-based ap- proach to decision making. IEEE Transactions on Systems, Man, and Cybernetics, Special Issue, Part C 33(1):102–113. Tien, J. M., and D. Berg. 1995. Systems engineering in the growing service economy. IEEE Transactions on Systems, Man, and Cybernetics Part C 25(5):321–326. _____. 2003. A case for service systems engineering. Journal of Systems Science and Systems Engineering 12(1):13–38. _____. 2006. On services research and education. Journal of Systems Science and Systems Engineering 15(3):257–283. _____. 2007. A calculus for services innovation. Journal of Systems Science and Systems En- gineering 16(2):129–165. Tien, J. M., and M. F. Cahn. 1981. An ealuation of the Wilmington Management of Demand Program. Washington, DC: National Institute of Justice. Tien, J. M., A. Krishnamurthy, and A. Yasar. 2004. Towards real-time customized manage- ment of supply and demand chains. Journal of Systems Science and Systems Engineering 13(3):257–278. Yasar, A. 2005. Real-time and simultaneous management of supply and demand chains. Ph.D. thesis. Troy, NY: Rensselaer Polytechnic Institute. Zadeh, L. A. 1996. The evolution of systems analysis and control: A personal perspective. IEEE Control Systems Magazine 16(3):95–98.