Most improvement efforts have an element of divide and rule to them; our hope is that by carving up the territory, we can conquer it. But any top-level representation of a complex system is, by it's nature, just an abstraction. As such, it will highlight some aspects of the system, and suppresses visibility of others. We do this in order to communicate about certain behaviors or features of the system, usually by telling stories about those parts (though we often don't write those stories down). There can thus be many different top-level views of any system. It is possible for them all to be concurrently valid, useful, and consistent with one another, but this doesn't just happen automatically; it must be designed to be so.

Think of the human body as a system. You could chose to represent that system to emphasize many different underlying aspects (skeletal, nervous, circulatory, digestive, respiratory, ), but trying to put all those details on the same diagram, or show their interdependencies, could be very confusing. Instead, in representations of the human body, multiple views of the same 'system' are usually utilized at different times, according to what one needs to communicate about, and to whom.

The medical care system itself is much the same; consider hospitals. Since there are many possible ways to represent the overall flow of material, transactions, or customers through such a system, you could represent the overall situation as a series of organizational hand-offs of responsibility, as a sequence of time-sliced views, or as transactional states according to the processing of different types of work through the system. What's important, in picking one or more such top-level views, is to decide what the purpose of that view is for, and ensure that if you try to use it for something differently than it was originally designed for, that you adopt a different view for that purpose. You could carve up the work as a value stream (with patients passing through emergency care, day surgery, or long-term care, each traversing different paths). You could also choose to visualize them as a collection of services (Orthopedics, Neurology, Cardiology, Radiology, General Surgery, ...), each with different patient responsibilities. You could also look at the flow from the perspective of users of consumables (Admissions, Nursing, Operating rooms, Laboratories, Central stores), or alternatively from the perspective of managing a series of projects (for example, transitioning to a new medical records system), or as a timeline (how is the Monday experience different from the Friday experience?).

It is common for organizations to attempt to craft a single, dominant view of the underlying processes in an organization. When this is attempted, it is often to emphasize the decomposition that the groups making the decomposition feel are most meaningful to them; such a view is likely to be based upon their own perceptions of how the organization works. To address systematic improvements successfully, the goal should not be to craft top-level structures for the purposes of representing underlying organizational structures (which can evolve with time), or seeking an upper hand in power or politics. Instead, it should be organized to exploit best practices in process design, gain insights about the underlying system, and understand interactions of the components under operation in that system, so that the system can be improved. Deciding how to 'slice and dice' that system can thus be a quite important decision.

At the most basic level, medical care can be considered to be comprised of a series of macro-process steps highlighted on the diagram above. Recognize that at this level, one cannot really tell if we're processing human patients, animals, or cars in a repair facility; those details all are revealed at lower levels in the process architecture. But the choices at the top level of what to represent in this view are still important, to highlight the primary actions and decision-making steps associated with the patient's (or car's, or animal's) processing over time.

The medical community has recognized that decision-making is at the core of all systematic improvements, and have crafted their steps accordingly, so that their macro steps and associated mental model are understood, accepted, and supported across all their services (organizations), whether in a hospital, in medical schools, in research institutions, or in communities of practice. Since each of these individual steps - diagnosis, work-up, treatment, and follow-up - have well-defined starting and ending points, and clearly defined expectations for information hand-offs, this view can then be used to track how all work proceeds, as it moves through the system, regardless of what type of patient is that is being seen, or which services are involved in their course of care. By adopting such a view, they have been able to ask questions like "how frequently are our diagnoses correct?', or 'how often has our work-up adequately prepared patients for treatment?'. To answer questions such as these, the chosen view also implies that there are certain underlying assumptions that must be satisfied.

For example, in order for this to work at lower levels, the definition of what constitutes a 'patient' and 'types of diseases' must remain constant throughout processing. Consider an alternative view, shown on the left, that could be crafted for processing patients. We could decide that we only wanted to focus on the cases that successfully navigated their way through these five steps, and not capture or seek to understand the reasons for failure. In this alternative representation, it would not make any sense to discuss the quality of planning, since by definition, under this alternative mental model, the underlying system concept is one that performs best-effort planning; not being able to successfully plan everything under this alternative paradigm is acceptable, and therefore, analyzing the reasons for being unable to plan might be impractical (perhaps due to constraints on resources). There are situations in which this alternative concept may make business sense; for example, if you had a very expensive Computed tomography scanner that you had purchased and that you had to recover costs for, you might want to maximize the flow of patients that you pushed through it, so you could be sure to have the greatest caseload to distribute those costs across. In this case, such a design might make a lot of sense to the hospital administration (but not necessarily to the patients!). But if your goal was to determine how effective your CT scanner was in improving patient outcomes, such a design would be clearly less useful.

In practice, it's been determined that such CT scans can generate a lot of expensive, but unnecessary 'false positive' findings. These findings then have to be acted on through other means, and thus can drive up patient costs unnecessarily. This highlights how easily sub-optimization can occur if the wrong top-level view is provided, and helps to emphasize the importance of getting your 'top-level' design right.

This top-level design is critical in communicating to affected stakeholders what the important strategic leverage points are for change. However, since they are just high-level depictions of work, they will not by themselves be sufficient to help all stakeholders understand what such changes actually mean to their own work, or what they must do to support these changes. Unfortunately, if they believe such changes will make their work more difficult (whether they really will or not), they will still provide a basis of conflict and resistance, even though in practice, no one really may have an adequate foundation to determine what is really going on. For example, isolating a problem to its underlying root cause can be very difficult, whether you are doing that for a patient or a process. When the capability of the process hasn't yet been characterized, or if the organization's maturity in utilizing it has not yet been determined, you may just be groping in the dark.

Flow time through a system is often critically important, especially in medicine. Studies show us time and again that if you walk into an emergency room with chest pains, there's a 90-minute window in which you must receive the appropriate treatment, if you are to get the best outcome. This is similar to the learning that the military has incorporated regarding processing time requirements for wounded soldiers, which has resulted in cuttting the death rate in half in the Iraq War. Improvements to overall flow can thus be enormously important. Showing how this can be done, and helping instill a sense of will and belief in the transformation, are key accelerators, and should be a part of the dialog with front-line people; such changes cannot just be pushed from the top. But to improve such flows, the actual flow, rather than the ideal flow, must be confronted. And to do that, rework must be considered.

Systems must be designed for robustness in both normal and non-normal operation; indeed, it is the performance of a system under load or under less than optimal conditions that often determines whether the system is successful or not. Complications are frequent in execution, and can be caused by problems with the process, training, execution, or inherent problems with the patient itself. In the latest version of our diagram, the original representation has been redrawn to highlight that work itself often has to be redone because of such complications. For example, in practice, diagnoses are only right about 90% of the time. Some of these are caught during work-up, and others are not discovered until treatments actually begin. Slips (failures in execution) are injected at one point in the overall flow, and are detected at a later step, and the longer you go, the more rework may have resulted.

When time matters, the costs can be even greater than the costs of the rework itself. Yet the further you go until such such defects are discovered, the more time that is wasted, and the greater the scrap and rework that there may be. For example, the flow (3)->(4)->(4)->(3)->( 3)->(4)->(5)->(7)->(3)->(4)->(5)->(6) is clearly longer than (3)->(4)->(5)->(6). As such time flow problems are encountered, there is the added complication that people begin to work under heavier loads, and thus rush what they do, which can make them even more error-prone; since more people are involved, it also increases the probability that errors in communications will occur. The system attempts to adopt, but unless it is designed to manage this flow, the flow through the system will begin to be 'choked' at bottlenecks, and it will not always be obvious where those bottlenecks are (since they depend upon what the pattern of successive work through the system as been, over time).

One might think that one way of reducing costs would thus be to focus on reducing the 'cycle time' of patients in hospitals, since hospital care itself is generally more expensive than outpatient care. While this has been attempted, it has not offered much cost leverage; this is because nearly all the costs of care in a hospital occur in the initial days of care. Studies have shown that you can cut a hospital stay by as much as 25%, but this will only reduce total costs by about 3%; cycle time reductions will not result in linear cost improvements. Therefore, as in most systems, sustainable cost reductions only are achieved by managing the quality and efficiency of work and information exchanges within the system itself. You have to determine how well each component of the system is working, and design new approaches for performing that work. And you must understand the variety of work, and tailor your solutions to accommodate that variety, or eliminate that variation.

Measurement is an essential ingredient to all of this decision-making, and it's addition in this further refinement of our diagram reveals an important point. Until the context in which measurements are made is established, and until the underlying mechanisms for collecting, analyzing, and reporting measurements are themselves reliable, the use of such measurements throughout the system itself can introduce rework and risk producing undesirable outcomes. Measurements are not isolated collections of data, but must be the integrated focus of each step in the overall effort. Measurements cannot be a quick afterthought, but must be designed along with the rest of the system. And this design should be focused on answering specific questions that lead towards the achievement of defined objectives. Otherwise, one is measuring for the sake of measuring, which is both costly and likely to be misleading.

Once a foundation of measurement is established, there also needs to be a means of coordinating the overall process, as depicted with the addition shown on the diagram on the right. While workflow techniques can be used to track and record progress in performing these activities over time, their interactions and collections must be integrated into the rest of the steps, just as measurements are. Regardless of the underlying technologies available to perform this tracking, someone needs to be responsible both for the required individual decisions, and for the overall outcome, throughout the processing.

Education and mentoring also play a critical role here, as there must be the means to guide the work and assure that the process is followed. In How Doctors Think, Jerome Groopman describes some of the challenges that this raises:

My generation was never explicitly taught how to think as clinicians. We learned medicine catch-as-catch-can. Trainees observed senior physicians the way apprentices observed master craftsmen in a medieval guild, and somehow the novices were supposed to assimilate their elder's approach to diagnosis and treatment. Rarely did an attending physician actually explain the mental steps that led him to his decisions. Over the past few years, there has been a sharp reaction against this catch-as-catch-can approach. To establish a more organized structure, medical students and residents are being taught to follow preset algorithms and practice guidelines in the form of decision trees. This method is also being touted by certain administrators to senior staff in many hospitals in the United States and Europe. Insurance companies have found it particularly attractive in deciding whether to approve the use of certain diagnostic tests or treatments.

The trunk of the clinical decision tree is a patient's major symptom or laboratory result, contained within a box. Arrows branch from the first box to other boxes. For example, a common symptom like "sore throat" would begin the algorithm, followed by a series of branches with "yes" or "no" questions about associated symptoms. Is there a fever or not? Are swollen lymph nodes associated with the sore throat? Have other family members suffered from this symptom? Similarly, a laboratory test like a throat culture for bacteria would appear farther down the trunk of the tree, with branches based on "yes" or "no" answers to the results of the culture. Ultimately, following the branches to the end should lead to the correct diagnosis and therapy.

Clinical algorithms can be useful for run-of-the-mill diagnosis and treatment - distinguishing strep throat from viral pharyngitis, for example. But they quickly fall apart when a doctor needs to think outside their boxes, when symptoms are vague, or multiple and confusing, or when test results are inexact. In such cases - the kinds of cases where we most need a discerning doctor - algorithms discourage physicians from thinking independently and creatively. Instead of expanding a doctor's thinking, they can constrain it.

There is an obvious tension here between the use of professional intuition and the application of detailed procedures to a given situation. If a procedure is too detailed, innovation may be stifled; if it is not detailed enough, variations in results may be due to uncontrolled aspects in how an intervention is delivered, rather than what the intervention itself dictates. Additionally, within the medical community, as in many other professionals, a belief in personal responsibility is deeply ingrained in the professional culture, and is responsible for their fierce attachment to individual autonomy.

Case management, by itself, can only be meaningfully performed if there is a historical record to track results over time. Such record keeping provides an essential communications across caregivers, whether care is performed within one hospital unit, or as a patient is transferred to other units. Until there is thorough and disciplined record-keeping, this communications itself can become a significant source of problems, as individuals with different 'mental models' of what is going on inject different strategies in the case management.

A case may be managed by many different services within a hospital over time. Imagine how crazy it would be if each of these services had their own version of the patient's history. Different groups can certainly use different views of information, much as we have shown that different views of the same patient can be maintained. But the underlying information systems (both written and IT-based) that support these views must be properly designed to allow this. When this is the case, these different views can be generated by making different queries to the same information database. Unfortunately, in many businesses, it's unfortunately often been the case that no one has had that design authority, or the time to transform the organization to input and output data from such a database, and so instead, a series of independent, and unreconciled views may all be in use, all derived from slightly different data sources. Until reconciliation of this information happens, an organization can spend an inordinate amount of time in confusion and debates about what is really going on, or should happen.

While quite a bit of work has been done to automate record-keeping within hospitals, it is only just now beginning to be possible across hospitals; previous research typically required retrospective chart reviews of selected patients, which was time-consuming, error-prone, and selectively done. One of the most important questions to ask about such views is thus whether there is a desire to recognize and act on patterns across sites or not, and determine how much effort should be expended in order to integrate, or retrospectively reconcile, these views.

Once a commitment has been made to adopt a 'single source of patient data', an entire branch of medicine (Epidemiology) studies the factors which affect the health and illness of a given population, in order to manage the associated risks, determine the best course of treatment, and improve the underlying approaches to both. This ultimately rests on the management of this information, the effective flow of patients through the system, and effective decision-making along the way.

Within this macro process context, many different communities of practice operate (across all healthcare occupations, but especially within nursing and medical specialties). The levels of guidance for each vary according to the necessary skills and training required of their various members to fulfill their respective roles and duties. Documented, low-level procedures exist for key detailed steps that are common across these communities of practice - for example, when calibrating equipment, or preparing and setting up the required equipment, parts, and tools for a medical procedure. Well defined protocols are established when these communities of practice must cooperate towards some collective goal, such as when communicating status about a population of patients as shift changes occur daily, scheduling key resources like operating rooms, or when orchestrating organ transplants across hospitals. Knowledge management within these communities of practice is largely accomplished through two mechanisms:

• well-defined medical education for the profession that builds on a series of stages (pre-medical, medical school, internship, residency, fellowship, and continuing education)
• extensive medical literature, which describes new methods and shares data collected from applying those methods against a controlled population; each of these is essential in communicating and coordinating the introduction of new techniques, treatments, and technologies throughout a broad population of practitioners.

Through all of this, there is a common body of knowledge, and many different levels of understanding required for two key topics, according to each practicing specialty: physiology, or the functions of the patient and their interactions, and diseases, or what can go wrong with those functions, and how that will manifest itself over time. Applying this knowledge relies on the developed skills of individuals; the extent of underlying formal process guidance, in a traditional, Six Sigma business sense, is quite limited, though the level of training (both classroom and on-the-job) is quite considerable. This works because of the professionalism of the underlying communities of practice, who self-police the preparation and oversight of their respective members, typically under a philosophy of 'See one, Do one, Teach one'; this of course requires careful mentoring by other, more qualified members within the community, and coordinated certification and licensure so that these underlying skills can be demonstrated and assured over time, as the populations within these communities of practice migrate and evolve.

Just as health care professionals have a culture which must be addressed within this system, so do the consumers of these services. The decision-making these consumers make is equally important in designing the health care system. Our final addition acknowledges this important wrinkle, and highlights that consumers have choices. Their choices, made outside of the health care system itself, often determine how successful interventions within the system itself will be.

The role played by the General practitioner, as gatekeeper into the hospital system, is equally important, but is unfortunately often not able to be engaged, if patients chose to only enter into the system through the emergency department. Individuals chose to go to their physician when they are sick, and aren't big on periodic health checkups. Unless the family practice physician is effective in gatekeeping and proactively managing their own portfolio, the flow of patients into the hospital will be more or less effective. A recent trend towards wellness care is industry's attempt at accomplishing their own interventions, given this situation, but obviously requires another culture change on the part of an even broader customer community.