Controlling the Complex Adaptive System

The most alert and knowledgeable readers will have already noticed a contradiction in the title. While “control” is a traditional engineering concept and focus, its meaning in the realm of complexity can best be described as “eroded.” Even that is probably a generous description. A more accurate title statement would likely be “Influencing the Complex Adaptive System.” But, the contradiction sets the tone for what follows here—an example of the folly of attempting to establish “control” over the complex adaptive system.

We should begin with a description of the nature of the complex adaptive system. It is that very nature that creates impediments to control.

Complex adaptive systems can be described as an arrangement of individual elements (called “agents” in the study of complexity) with freedom to act in ways that are not necessarily predictable. The actions of the system agents impact the system context for other elements. That means that the agents “learn” to adapt based on their interactions with the other agents of the systems. There is little chance that the complex adaptive system will reach an equilibrium state where the agents cease to adapt. This makes it extremely hard to predict, much less control, the emergent system behavior.

Classically, system control is achieved by the identification and alteration of the “root causes” and/or stimuli for the system behaviors. But, the causal chains from root cause through immediate causes in a complex adaptive system are continually changing as the system elements adapt to each other through their interactions. All this taken together means that the concept of linear cause and effect disappears quickly in a complex adaptive system, a circumstance which, therefore, brings about the derogation or outright demise of control.

The complex adaptive system looms large in the 21st century world. In the webs of systems of systems—particularly socio-technical systems—it is increasingly hard to find to find any which are not, or do not involve, complex adaptive systems. That means that any attempt to modify or create system solutions will affect or be affected by complex adaptive systems.

The difficulty of attempting to control a complex adaptive system can be illustrated with an example from healthcare. This example deals with a significant problem in healthcare delivery—the flow of patients through the healthcare providers’ care systems.

In response to the problem of long patient stays in the Emergency Departments (EDs) of British hospitals, the National Health Service of England instituted a “target” of four hours as a maximum ED stay. The application of the target was rolled out beginning in 2004, and by 2009, approximately 97% of the ED patients in England were leaving the ED in four hours or less to an apparent interventional success.

There were, however, some numbers behind the numbers that told quite a different story. In a study of the performance of EDs to their targets, an interesting trend was observed. (The study, Time Patients Spend in the Emergency Department: England’s 4-Hour Rule—A Case of Hitting the Target but Missing the Point? by Suzanne Mason, Ellen Weber, Joanne Coster, Jennifer Freeman, and Thomas Locker, can be found on Science Direct, as published in Annals of Emergency Medicine in May of 2012.)

The study of the British attempt at controlling ED patient flow, or “throughput” as it is known in that world, described the problematic effects of the targets. “Whether a patient stays 2 hours or 3 hours and 59 minutes is equally rewarded, and once the patient ‘breaches’ [the 4-hour stay], their length of stay becomes irrelevant (to the target at least).” Instead of improved ED processes resulting in better patient flows, the ratio of ED patients admitted to the hospital to those discharged increased, and the portion of patients moved out of the ED in the last 20 minutes of the 4-hour target became increasingly significant. As the study put it, “We hoped that the target would have led to improved processes, resulting in patients being treated sooner and leaving earlier across the four hours, without diminishing time for physician-patient interactions and care. We did not observe this pattern.”

The reason for this rests in the very nature of the ED administration as a complex adaptive system. The study showed that the EDs were meeting the targets. But this was accomplished through a variety of adaptive strategies. In writing about the application of targets in this case, the Institute for Healthcare Improvement (IHI) (Guiding the Flock: 3 Simple Rules to Improve Hospital-wide Patient Flow, IHI Staff, 2018) cited W. Edwards Deming. In The New Economics, Deming described three ways to meet a target without system improvement:

  1. Redefine the terms—In the ED example, this could involve admitting patients to hallway spaces instead of designated unit beds.
  2. Distortion and faking—In patient flow, this could involve submitting inaccurate data.
  3. Run-up costs—For example, units could add unnecessary beds, staff, or other resources.

Deming is pointing out that, especially in a complex adaptive system, the system agents have a number of adaptations open to them when targets are chosen. They will likely figure out how to meet the target without necessarily choosing to accomplish the more difficult purpose behind it.

In the EDs in the study discussed above, the staff found a variety of ways to move patients out of their domain (increased admissions which transferred responsibility for patient care to other departments, deferring the decisions until near the deadline and choosing other alternatives, etc.) thereby meeting the imposed standard. The target, imposed with the intention of improving patient care, did not have the intended effect. This was due to the failure of the target authors to consider the nature of the complex adaptive system into which it was introduced. The system agents (ED staff and administrators) adapted their behaviors to the change in the environment introduced by changed expectations, but did so without being constrained to accomplish the purpose of the target imposition.

In contrast to the Health Service approach, the IHI posited a more thoughtful approach that considers the ways that the system can adapt. They propose three simple rules to accomplish the same purpose as was intended in the English target. The “rules”—Right Care, Right Place; Right Time; and Operational Capacity—are intended not for imposition as a standard for judgment, but as guiding principles by which the system agents can learn and adapt their interactions and results. By recognizing and encouraging thoughtful adaptation, these rules or guiding principles increase the probability of productive adoption. What IHI proposes is aimed at influencing the system rather than a futile attempt at controlling it.

The wisdom in targeting influence rather than control has a message for all of us in our increasingly complex problem and solution spaces. Where the ability to control is eroded by complex adaptive systems, the better strategy is to influence the system. Careful consideration of how to influence the independent adaptation of system agents is critical to success in designing and modifying such systems.

Leave a Reply