The Three Mile Island nuclear power plant accident prompted the idea of Normal Accident Theory. This theory holds that as systems become complex, accidents become inevitable or normal (St. Pierre, et al., 2008). The susceptibility to accidents in these complex organizations is determined by the dimensions of interactive complexity and coupling (St. Pierre, et al., 2008). An accident is defined as “an incident in which non-trivial loss occurs” (Cooke & Rohleder, 2006, p. 214). An incident is an unexpected or unwanted change in process that has the potential to cause a loss (Cooke & Rohleder, 2006). An accident is classified as a disaster when loss of life or extensive property damage or money loss occurs (Cooke & Rohleder, 2006)…..
According to Tamuz and Harrison (2006), coupling refers to the “degree of dependence among system components” (p. 1658). Coupling can be described as tight or loose depending on the degree of interdependency (St. Pierre, et al., 2008). When system components are tight, changes in one system have major impacts on another. For example * a shutdown of a hospital’s server would shut down all operations that depended on web access such as email, web-based EMRs, etc. In a loose coupling the systems are more independent from each other or there are buffers in place to prevent destabilization (St. Pierre, et al., 2008).
Interactive complexity refers to the degree to which connections between the components are unexpected, unplanned or not visible (Tamuz & Harrison, 2006). In general, hospital systems tend to be loosely coupled (Tamuz & Harrison, 2006) with various professions and departments pursuing their own goals and outcomes, but the interactive complexity is high. An example would be a patient who was wrongly identified in an x-ray leading to misdiagnosis. The radiology and the MD function in separate silos and the wrong identification would not be expected or even visible until an intervention or lack of intervention caused a change in patient condition. The structural tenets of Normal Accident Theory have implications for hospitals who seek who reduce accidents and disasters.
Safety Dog's Blog 
Does interactive complexity deal with the impact of design of the environment upon actions and outcomes?
Absolutely! Interactively complex systems often have many branching paths among subsystems. Interactions between these can create unexpected failures. Often these interactions are unseen until something happens.
For the example of xray failure as mentioned above, an environmental design might be to have a small radiology department attached to a high risk area like a breast health center to create closer interaction between the departments. They would still be separate (loosely coupled) but designing a physical environment and communication process between them could reduce interactive complexity (or make the interaction more linear) and they would achieve their desired outcomes of providing accurate diagnoses of breast health issues.
For more information that might provide a better answer to your question:
This is a link to the full text of Tamuz and Harrison’s article.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1955347/
They provide an excellent explanation of the perspectives on healthcare safety from the Normal Accident Theory (NAT) perspective and the High Reliability Theory (HRT) perspective. At first these two theories were viewed as competing approaches but now many see them as complimentary frameworks for patient safety (Tamuz & Harrison, 2006).
According to NAT, the most dangerous situation is supposed to be having high interactive complexity and tight coupling. In tight coupling processes are so entwined that failure in one leads to a domino effect and multiple failures. In loosely coupled systems, the processes have more independence however interactive complexity can still exist and contribute to failures. Changes in one unit may seem to be independent of another unit until the hidden interactions surface.
An interesting example is found in Tamuz and Harrison (2006). CPOE reduces variance in order writing which has been shown to decrease errors by individuals. We may have isolated errors when one MD writes a wrong order. But it’s loose because all the new orders are independent of each other. CPOE reduces interactive complexity but actually creates a tight coupling. So if by human error something is entered incorrectly in the programming code, we now have every MD using the system making the error as opposed to the single individual. There is potential for massive failure.
The safest system would be one with loose coupling and predictable, visible, interactions. LEAN is one technique that is geared to reducing interactive complexities.