Management of complex adaptive systems requires

leadership rather than power, incentives and inhibitions

rather than command and control.

Health Care as a Complex

Adaptive System: Implications
for Design and Management

William B. Rouse

For several years, the National Academies has been engaged in a systemic
study of the quality and cost of health care in the United States (IOM, 2000,
2001; National Academy of Engineering and Institute of Medicine, 2005).
Clearly, substantial improvements in the delivery of health care are needed
and, many have argued, achievable, via value-based competition (e.g., Por-
ter and Teisberg, 2006). Of course, it should be kept in mind that our health
William B. Rouse is executive
care system did not get the way it is overnight (Stevens et al., 2006).
director, Tennenbaum Institute, Many studies by the National Academies and others have concluded that
Georgia Institute of Technology, a major problem with the health care system is that it is not really a system.
In this article, I elaborate on the differences between traditional systems and
and an NAE member. This article
complex adaptive systems (like health care) and the implications of those
is based on a presentation at differences for system design and management.
the NAE Annual Meeting Technical
Complex Adaptive Systems
Symposium on October 1, 2007.
Many people think of systems in terms of exemplars, ranging from vehi-
cles (e.g., airplanes) to process plants (e.g., utilities) to infrastructure (e.g.,
airports) to enterprises (e.g., Wal-Mart). In addition, they often think of
improving a system by decomposing the overall system performance and
management into component elements (e.g., propulsion, suspension, elec-
tronics) and subsequently recomposing it by integrating the designed solu-
tion for each element into an overall system design.
 The
18 BRIDGE

This approach of hierarchical decomposition (Rouse, emergent behaviors may range from valuable inno-
2003) has worked well for designing automobiles, vations to unfortunate accidents.
highways, laptops, cell phones, and retail systems that
• There is no single point(s) of control. System behaviors
enable us to buy products from anywhere in the world
are often unpredictable and uncontrollable, and no
at attractive prices. The success of traditional systems
one is “in charge.” Consequently, the behaviors of
depends on being able to decompose and recompose the
complex adaptive systems can usually be more easily
elements of the system and, most important, on some-
influenced than controlled.
one or some entity having the authority and resources
to design the system. Before elaborating on these characteristics in the con-
text of health care, it is useful to reflect on an overall
implication for systems with these characteristics. One
cannot command or force such systems to comply with
Hierarchical decomposition behavioral and performance dictates using any conven-
tional means. Agents in complex adaptive systems are
does not work for complex sufficiently intelligent to game the system, find “work-
adaptive systems. arounds,” and creatively identify ways to serve their
own interests.

The Health Care Game

Not all system design and management problems Consider the large number of players, or “agents,”
can be addressed through hierarchical decomposition. involved in the health care game (Table 1). It is reason-
For example, decomposition may result in the loss of able to assume that each type of agent attempts to both
important information about interactions among the serve its own interests and provide quality products and
phenomena of interest. Another fundamental problem services to its customers. However, there are conflict-
for very complex systems like health care is that no one ing interests among stakeholders, just as there are dif-
is “in charge,” no one has the authority or resources to ferent definitions of quality. Thus, even assuming that
design the system. Complex adaptive systems tend to all agents are well intentioned, the value provided by
have these design and management limitations. the health care system is much lower than it might be,
Complex adaptive systems can be defined in terms of in the sense that health outcomes may be compromised
the following characteristics (Rouse, 2000): and/or the costs of delivering these outcomes may be
excessive.
• They are nonlinear and dynamic and do not inherently
Working with the American Cancer Society, we stud-
reach fixed-equilibrium points. As a result, system
ied the value chain associated with disease detection
behaviors may appear to be random or chaotic.
(Rouse, 2000). Many people naively believe that new
• They are composed of independent agents whose detection technology is the key to successful detection.
behavior is based on physical, psychological, or social However, unless we address consumer awareness, con-
rules rather than the demands of system dynamics. sumer education, physician education, and consumer
advocacy, to name a few of the other components of the
• Because agents’ needs or desires, reflected in their rules,
value chain, patients may not experience the benefits
are not homogeneous, their goals and behaviors are likely
of new detection technologies. In general, enormous
to conflict. In response to these conflicts or competi-
investments in medical research will not substantially
tions, agents tend to adapt to each other’s behaviors.
improve health care outcomes unless they are intro-
• Agents are intelligent. As they experiment and gain duced with an understanding of the overall system.
experience, agents learn and change their behaviors In this context, it is useful to look more closely at
accordingly. Thus overall system behavior inherently the two cells in Table 1 that include physicians. One
changes over time. aspect of the overall health care value chain is the
process of education and certification that provides
• Adaptation and learning tend to result in self-
trained, licensed physicians. Physician education and
organization. Behavior patterns emerge rather than
training are currently being reexamined to identify
being designed into the system. The nature of
Spring 2008 19

TABLE 1 Stakeholders and Interests in Health Care

Stakeholder Risk Management Prevention Detection Treatment

Public e.g., buy insurance e.g., stop smoking e.g., get screened

Delivery System Cliniciansa Clinicians and providersb

Government Medicare, Medicaid, NIH, Government CDC, NIH, Government CDC, NIH, Government CDC,
Congress DoD, et al. DoD, et al. DoD, et al.

Non-Profits American Cancer American Cancer American Cancer

Society, American Heart Society, American Heart Society, American Heart
Association, et al. Association, et al. Association, et al.

Academia Business schools Basic science disciplines Technology and medical Medical schools
schools

Business Employers, insurance Guidant, Medtronic, et al. Lilly, Merck, Pfizer, et al.
companies, HMOs
a The category of clinicians includes physicians, nurses, and other health care professionals.
b The category of providers includes hospitals, clinics, nursing homes, and many other types of testing and treatment facilities.

future physician competencies and determine the best stakeholders and interests, layered by organization,
way to provide them. Some of the many stakeholders specialty, state, and so on. If this system is approached
in this process are listed below: in the traditional way, decomposing the elements of
the system, designing how each element should func-
• Accreditation Council for Continuing Medical
tion, and recomposing the overall system would be
Education
overwhelming. Thus we must address health care in
• Accreditation Council for Graduate Medical a different way and from a different point of view—as
Education a complex adaptive system.
• American Academy of Family Physicians Modeling Complexity
• American Board of Medical Specialties The first consideration in designing an effective
health care system is complexity. Figure 1 provides
• American Medical Association
a high-level view of the overall health care delivery
• American Osteopathic Association (AOA) network based on recent studies of service value net-
works (Basole and Rouse, 2008). Note that each node
• AOA Council on Postdoctoral Training
in the network includes many companies and other
• Council of Medical Specialty Societies types of enterprises.
Assessing the complexity of networks involves defin-
• Federation of State Medical Boards
ing the state of the network, that is, the identity of the
• Joint Commission on Accreditation of Healthcare nodes participating in any given consumer (patient)
Organizations transaction. We then use information theory to cal-
culate the number of binary questions that have to be
• Liaison Committee on Medical Education
asked to determine the state of the network. Given esti-
This list is representative, but not exhaustive. In mates of the conditional probabilities of a node being
addition, many functions of these organizations are involved in a transaction, complexity can be calculated
state specific, so there might be 50 instances of these and expressed in terms of binary digits (bits).
academies, boards, committees, and councils. Figure 1 summarizes an assessment of the complexity
Even from this brief description, it is apparent that of five markets. Note that the complexity of health care
the system of health care delivery involves what we is assessed to be 27 bits. This means that determin-
might call networks of networks or systems of systems ing which nodes (i.e., enterprises) are involved in any
that involve an enormous number of independent particular health care transaction would require on the
 The
20 BRIDGE

Government and
Policy Makers

Health
Insurance

Pharmacy

Pharmaceuticals Health
Wholesalers
Health Consumers
Providers
Medical
Equipment
R&D Laboratories

Other
Equipment

Figure 1 A summary of the complexity of five markets in the health care delivery network.

order of 1 billion binary questions. Thus it would be The idea of consumer-directed health care, however,
an enormous task to determine the state of the overall is going in the opposite direction in that it increases
health care system. complexity for consumers, and possibly for clinicians.
Notice the ratios of consumer complexity to total Using other markets as benchmarks, we would expect
complexity in Figure 2. Even though the retail market this push to fail, or at least to have limited success. Thus
is the most complex market, the consumer only has to the goal should be to increase the complexity of health
address a small portion of this complexity. The retail care where it can be managed in order to reduce com-
industry has been quite successful in managing the com- plexity for patients, their families, physicians, nurses,
plexity of bringing a rich variety of products and services and other clinicians.
to market without consumers having to be concerned The case for decreasing complexity for clinicians is
about how this cornucopia arrives on store shelves—or supported in the analyses by Ball and Bierstock (2007),
online outlets. who argue that enabling technologies should support
The telecom industry has the worst ratio, as anyone both clinicians’ workflow and “thought flow.” As long
who has tried to call for vendor technical support for as systems increase clinicians’ workload while providing
a laptop can attest. As a consumer, you need to know them few if any benefits, the adoption of technology
much more than you want to know about the hardware will continue to be difficult.
and software inside your laptop. A substantial portion
of innovations being pursued in this market are aimed Design Implications
at significantly reducing the complexity experienced by Our studies of the complexities of markets have led
consumers. We expect that those who are most suc- to two propositions for which we have found consider-
cessful at reducing consumer complexity will be the able supporting evidence. Thus we now feel they can
winners in this rapidly changing market. be articulated as design principles.
Spring 2008 21

35

Consumer Total
30

25
Complexity (Bits)

20

15

10

0
Aerospace Automotive Retail Health Care Telecom

Figure 2 Comparative levels of complexity for five markets.

Design Principles whether they are customers, partners, collaborators,

First, the nature and extent of business-to-consumer channels, competitors, or regulators. Starting with
service value determines business-to-business service this model of the enterprise, the overarching strat-
value, as well as the value of products and other value egy should focus on increasing complexity where it
enablers. In the context of health care, the value pro- can be managed best and decreasing complexity for
vided to consumers and the payment received for this end users.
value determine the financial potential for all of the
Designing Agile Complex Systems
other players in the network. For example, if con­sumers
do not value and will not pay for a particular test or Most mature enterprises can manage design, develop-
treatment, none of the participants in this network will ment, manufacturing, and sustainment of products and
be rewarded, no matter how far upstream they are from services. Few enterprises can manage economies, mar-
patients and physicians. This is complicated, of course, kets, competitors, and end users. Put simply, because
when third parties pay for products and services. one cannot control the state of health, education, or
The second principle involves the ratio of business-to- preferences of those who seek health care, one cannot
consumer complexity to total-market complexity. This assume that they will be able and willing to manage the
ratio tends to decrease as markets mature. The most complexity of the system. Consequently, the design
successful players in a market are those who contribute should be focused on managing complexity by provid-
most to this decrease. To accomplish this, they usually ing ways of monitoring and influencing system state,
have to increase business-to-business complexity, which performance, and stakeholders, as elaborated below.
often increases total-market complexity. However, this This strategy for managing complexity can be facili-
is done in ways that decrease the ratio of consumer com- tated by also designing an agile enterprise that can
plexity to total complexity. This is the strategy that readily make decisions to redeploy resources to address
health care should pursue. opportunities and problems (Rouse, 2006). Achieving
As much as possible, complex systems should be agility requires trading off optimization to create the
designed and not just emerge. Design should begin leanest possible enterprise while maintaining flexible
with the recognition that the health care enterprise— resources that can respond to contingencies. The low-
as a system—includes all stakeholder organizations, est cost health care system would be quite fragile if these
 The
22 BRIDGE

contingencies have characteristics outside of the design Finally, value implies relevant, usable, and useful
assumptions for which the system has been optimized. outcomes, which require that stakeholders under-
Recent research indicates that the best way to address stand and appreciate the management philosophy of
this trade-off is to use the construct of system architec- the system and its implications. In a complex adap-
tures (Rouse, 2007a). tive system, a lack of understanding and/or appre-
ciation tends to result in “dysfunctional” behaviors
Management Implications by one or more stakeholder groups, although these
Complex adaptive systems can be designed, but only behaviors may be well intended and even reason-
to a certain degree. For instance, as outlined above, able according to the stakeholders’ understanding of
one can design an enterprise-wide information system the ends being sought and the means appropriate to
for such systems (Zammuto et al., 2007). However, achieving them.
these systems cannot be designed in the same sense that
a vehicle or industrial process can be designed. This is Organizational Behaviors
because complex adaptive systems have strong tenden- The best way to approach the management of com-
cies to learn, adapt, and self-organize. plex adaptive systems is with organizational behaviors
Consequently, the task of managing complex adap- that differ from the usual behaviors, such as adopting
tive systems becomes a challenge because, in effect, the a human-centered perspective that addresses the abili-
system keeps redesigning itself. In fact, the construct of ties, limitations, and inclinations of all stakeholders
“management” has to be viewed differently for complex (Table 2) (Rouse, 2007b).
adaptive systems than for other types of systems. Con- Given that no one is in charge of a complex adaptive
sider the management philosophy. Traditional systems system, the management approach should emphasize
are managed to minimize cost. Health care must be leadership rather than traditional management tech-
managed to maximize value. niques—influence rather than power. Because none,
or very few, of the stakeholder groups in the health
Value Philosophy care system are employees, command and control
Recent attempts at health care reform have tended, in has to be replaced with incentives and inhibitions.
effect, to pursue the lowest cost acceptable health care No one can require that stakeholders comply with
for our population. In contrast, we should be pursuing organizational dictates. They must have incentives
the highest value health care. Value focuses on orga- to behave appropriately.
nizational outputs (or outcomes), rather than inputs. Not only are most stakeholders in health care inde-
Thus we should emphasize the health states (outputs) of pendent agents, they are also beyond direct observa-
patients rather than the revenues (inputs) of providers. tion. Thus one cannot manage their activities but can
Value relates to the benefits of outcomes, rather than only assess the value of their outcomes. In a traditional
the outcomes themselves. From this perspective, we system, one might attempt to optimize efficiency.
should be very interested in productivity improvements However, the learning and adaptive characteristics
attributable to wellness, rather than simply the absence of a complex adaptive system should be leveraged to
of sickness. In an increas-
TABLE 2 Comparison of Organizational Behaviors
ingly knowledge-based
economy, the intellectual Traditional System Complex Adaptive System
assets embodied in people Roles Management Leadership
are central to global com-
petitiveness and economic Methods Command and control Incentives and inhibitions

growth. A recent report Measurement Activities Outcomes

from the Milken Institute Focus Efficiency Agility
shows that the costs of lost
productivity are often four Relationships Contractual Personal commitments
to five times greater than Network Hierarchy Heterarchy
the costs of health care
Design Organizational design Self-organization
(DeVol et al., 2007).
Spring 2008 23

encourage agility rather than throttled by optimization system. If we were discussing the banking system, the
focused on out-of-date requirements. answer would be the Federal Reserve Bank. The Fed
Of course, there are contractual commitments in com- does not tell banks what to do, but it sets the prime
plex systems, but because of the nature of these systems, interest rate and determines each bank’s reserve require-
stakeholders can easily change allegiances, at least at the ments. Banks and investors then decide how they want
end of their current contracts. Personal commitments, to adapt to any changes.
which can greatly diminish the risks of such behaviors, The health care system has no overseer,1 although
imply close relationships rather than arm’s-length rela- some have argued that there should be one, considering
tionships among stakeholder groups and transparent the importance of the health of the country’s human
organizational policies, practices, and outcomes. capital to competitiveness and economic growth. The
Work is done by heterarchies, whereas permissions question is which variables an overseer might adjust.
are granted and resources provided by hierarchies. To Perhaps it would adjust reimbursement rates in relation
the extent that the heterarchy has to stop and ask the to the value of health outcomes. Admittedly, outcomes
hierarchy for permission or resources, the efficiency and can be difficult to characterize and calibrate, and deter-
effectiveness of the system is undermined. Decision- mining attribution of causes of outcomes can be difficult
making authority and resources should be delegated because multiple actors are involved and outcomes only
to the heterarchy with, of course, the right incentives emerge over time. Nevertheless, at the very least, we
and inhibitions. should be able to characterize and assess bad outcomes
Finally, as noted throughout this article, because (IOM, 2000).
complex adaptive systems self-organize, no one can
impose an organizational design. Even if a design were
imposed, it would inevitably be morphed by stakeholders
as they learn and adapt to changing conditions. In that Because complex adaptive
case, the organization that management would think it
was running would not really exist. To the extent that
systems self-organize,
everyone agrees to pretend that it still exists, or ever no one can impose an
existed, value will be undermined.
organizational design.
Information Systems
Based on the organizational behaviors for complex
adaptive systems described above, information to over- More controversially, an oversight organization
see the system should include the following elements: might adjust tax rates so that (risk-adjusted) high-value
providers would pay lower taxes, perhaps reflecting the
• Measurements and projections of system state in
economic benefits of high-value health care. I know
terms of current and projected value flows, as well as
this idea is controversial because I have presented it to
current and projected problems.
various groups of thought leaders in health care. Beyond
• Measurements and projections of system performance the philosophical objection to using the tax system to
in terms of current and projected value, costs, and improve the public good, the most frequent criticism
metrics (e.g., value divided by cost), as well as current is that providers cannot fully affect health outcomes
and projected options for contingencies. because patient behaviors are also essential to success.
However, this is also true of markets of all kinds. The
• Observations of system stakeholders in terms of
enterprises that succeed are the ones that convince
the involvement and performance of each stake-
and incentivize consumer behaviors that co-create
holder group.
• Capabilities for measurement, modeling, and display 1 Some have argued that the Centers for Medicare and Medicaid Ser-
vices (CMS), an element of the U.S. Department of Health and Human
of system state, including agile “What If?” experimen- Services, plays a dominant role in setting reimbursement levels for
tation and adaptation. patients enrolled in these programs via the Medicare Payment Advisory
Commission. However, CMS does not oversee the whole health care
The question arises about who would be looking at system or address the overall health outcomes and economic impacts
discussed in this article.
and using the information for the whole health care
 The
24 BRIDGE

high-value outcomes. Success for the health care approaches 100 percent, many things would certainly
model depends on providers seeing themselves as change, and health care might be rationed at that point.
ensuring high-value outcomes, rather than being However, by acting long before we reach such a crisis
re­imbursed for the costs of their services. point, we can engineer much better solutions to provid-
ing high-value health care.
Conclusions Systems engineering for health care can operate on
The models and analyses discussed in this article can multiple time scales. Several of the ambitious ideas out-
be summarized in just two words—information and lined here will take several years, or more, to be fully
incentives. Substantial improvement in the system of realized. However, in the process of pursuing these
health care in the United States will require that stake- ideas, we will gain in understanding, which will inevi-
holders have easy access to information on the state and tably result in our identifying much low-hanging fruit,
performance of the whole system, or any subsystem, as that is, short-term opportunities that can be pursued
well as information on best practices at all levels. This much faster than the overarching vision. These short-
information would be used to assess current and emerg- term pursuits will undoubtedly improve the health care
ing situations in this complex adaptive system, which system, even as we work on the long-term vision to
would lead to adjustments of incentives and inhibitions transform the overall system.
to motivate stakeholders to change their behaviors to We need to analyze and design the systemic nature
continually increase value. of health care delivery and not continue to let it evolve
In general, incentives are essential to complex adap- and see whether one idea or another works. Complex
tive systems. Outcomes, as well as activities, must be adaptive systems require sophisticated and sometimes
incentivized. Payments to providers should reflect subtle analyses and designs, which will no doubt require
the (risk-adjusted) value of the outcomes achieved experts in a wide range of disciplines beyond engineer-
regardless of the cost incurred to achieve them. Poorly ing. However, a strong competency for analysis and
informed and/or out-of-date practices should be disin- design of complex adaptive systems will serve us well.
centivized. High-performing providers should reap sub-
stantial rewards, and poorly performing providers will Acknowledgements
go out of business. In this way, the average performance I am indebted to Marion Ball, Kenneth Boff, Dale
level will continually rise. Compton, Jerome Grossman, John Paul, and Stephen
Wellness, which contributes to productivity, should Schoenbaum for their comments and suggestions on
also be incentivized. Building on the recent report of drafts of this article.
the Milken Institute (DeVol et al., 2007), an economic
model could be developed of the relationship between References
wellness and productivity to provide a basis for determin- Aspinall, M.G., and R.G. Hamermesh. 2007. Realizing
ing how much should be invested in public awareness and the promise of personalized medicine. Harvard Business
education. The model would also be a basis for designing Review 85(10): 108–117.
tax incentives for employers who offer wellness programs Ball, M.J., and S. Bierstock. 2007. Clinician use of enabling
and whose employees participate in them. technology. Journal of Healthcare Information Manage-
These are “big” ideas that need much refinement, ment 21(3): 68–71.
analysis, and debate, and we must keep in mind the Basole, R., and W.B. Rouse. 2008. Complexity of service
inertial power of the status quo. A few years ago, The value networks: conceptualization and empirical investiga-
Economist (2004) published a long survey article on tion. IBM Systems Journal 47(1): 53–70.
health care financing, which admonished would-be DeVol, R., A. Bedroussian, A. Charuworn, A. Chatterjee,
reformers to remember that every health care dollar I. Kim, S. Kim, and K. Klowden. 2007. An Unhealthy
saved is somebody’s income. Aspinall and Hamermesh America: The Economic Burden of Chronic Disease. Santa
(2007) reinforced that idea in a recent assessment of the Monica, Calif.: Milken Institute.
promise of personalized health care. Economist. 2004. Survey of Healthcare Finance. July 15.
Thus we clearly need incentives for key stakeholders IOM (Institute of Medicine). 2000. To Err Is Human: Build-
to change. One incentive might be a crisis. For exam- ing a Safer Health System. Washington, D.C.: National
ple, when the percentage of GDP devoted to health care Academies Press.
Spring 2008 25

IOM. 2001. Crossing the Quality Chasm: A New Health Rouse, W.B. 2006. Agile Information Systems for Agile
System for the 21st Century. Washington, D.C.: National Decision Making. Chapter 2 in Agile Information Sys-
Academies Press. tems, edited by K.C. Desouza. New York: Butterworh-
National Academy of Engineering and Institute of Medi- Heinemann.
cine. 2005. Building a Better Delivery System: A New Rouse, W.B. 2007a. Complex engineered, organizational and
Engineering/Health Care Partnership. Washington, D.C.: natural systems: issues underlying the complexity of systems
National Academies Press. and fundamental research needed to address these issues.
Porter, M.E., and E.O. Teisberg. 2006. Redefining Health Systems Engineering 10(3): 260–271.
Care: Creating Value-Based Competition on Results. Bos- Rouse, W.B. 2007b. People and Organizations: Explorations
ton, Mass.: Harvard Business School Press. of Human-Centered Design. New York: John Wiley and
Rouse, W.B. 2000. Managing complexity: disease control as Sons.
a complex adaptive system. Information • Knowledge • Stevens, R.A., C.E. Rosenberg, and L.R. Burns, eds. 2006.
Systems Management 2(2): 143–165. History and Health Policy in the United States. New
Rouse, W.B. 2003. Engineering complex systems: implica- Brunswick, N.J.: Rutgers University Press.
tions for research in systems engineering. IEEE Transac- Zammuto, R.F., T.L. Griffith, A. Majchrzak, D.J. Dough-
tions on Systems, Man, and Cybernetics—Part C 33(2): erty, and S. Faraj. 2007. Information technology and the
154–156. changing fabric of organization. Organization Science
18(5): 749–762.