Prediction of Rate and Severity of Adverse Perioperative Outcomes: “Normal Accidents” Revisited

Authors


Abstract

The American Society of Anesthesiologists Physical Status classification system has been shown to predict the frequency of perioperative morbidity and mortality despite known subjectivity, inconsistent application, and exclusion of many perioperative confounding variables. The authors examined the relationship between the American Society of Anesthesiologists Physical Status and both the frequency and the severity of adverse events over a 10-year period in an academic anesthesiology practice. The American Society of Anesthesiologists Physical Status is predictive of not only the frequency of adverse perioperative events, but also the severity of adverse events. These nonlinear mathematical relationships can provide meaningful information on performance and risk. Calculated odds ratios allow discussion about individualized anesthesia risks based on the American Society of Anesthesiologists Physical Status because the added complexity of the surgical or diagnostic procedure, and other perioperative confounding variables, is indirectly factored into the Physical Status classification. The ability of the American Society of Anesthesiologists Physical Status to predict adverse outcome frequency and severity in a nonlinear relationship can be fully explained by applying the Normal Accident Theory, a well-known theory of system failure that relates the interactive complexity of system components to the frequency and the severity of system failures or adverse events. Mt Sinai J Med 79:46–55, 2012.© 2012 Mount Sinai School of Medicine

In 1941, Dr. Meyer Saklad, acting for an American Society of Anesthetists' committee of 3, published a paper in Anesthesiology titled “Grading of Patients for Surgical Procedures.”1 The committee had been asked to develop “a system for the collection and tabulation of statistical data in anesthesia.” The committee proposed a system of 6 categories, including 2 emergency categories along with definitions and examples. Saklad was careful to distinguish between “physical state” and “operative risk.” At the time it was common for an anesthesiologist to identify a patient as high risk or low risk or some gradient within those extremes. Saklad and his committee sought to eliminate all of the variables that might apply to operative risk, including the operation, individual physician skills, and other factors that might be used in making a judgment of risk. They believed that physical state could be defined with a commonly understood language that would eliminate all of the indeterminate variables associated with a subjective risk assessment.

PHYSICAL STATUS CLASSIFICATION SYSTEM OF AMERICAN SOCIETY OF ANESTHESIOLOGISTS

Two decades later, this system became the known as the American Society of Anesthesiologists (ASA) Physical Status (PS) classification system after being amended to include 5 categories and an E modifier for emergency cases.2 Relatively minor changes were made more recently, including an ASA PS 6 category for transplant donors diagnosed with brain death. The 1961 classification system was easy to apply and became widely used throughout the United States. Today, the ASA PS classification system (Table 1) is the most commonly used preoperative patient classification system throughout the world.

Table 1. American Society of Anesthesiologist Physical Status Classification System*
  • *

    The definitions in Table 1 appear in each annual edition of the American Society of Anesthesiologist Relative Value Guide.

ASA PS 1A normal healthy patient
ASA PS 2A patient with mild systemic disease
ASA PS 3A patient with severe systemic disease
ASA PS 4A patient with severe systemic disease that is a constant threat to life
ASA PS 5A moribund patient who is not expected to survive without the operation
ASA PS 6A declared brain-dead patient whose organs are being removed for donor purposes

Saklad's committee hoped that the use of the classification system, based on the patient's physical status, would eventually lead to identifiable associations with results.1 Predictably, reported associations generally showed that there were more adverse outcomes with higher rank order ASA PS categories.3–5 The authors have further shown that this relationship between the ASA PS and adverse outcome is nonlinear.6 This nonlinear relationship is worthy of study because it can provide meaningful information on performance and risk with applicability toward informed consent and the true valuation of surgical procedures.

Today, the physical status (PS) classification system of the American Society of Anesthesiologists (ASA) is the most commonly used preoperative patient classification system throughout the world.

NORMAL ACCIDENT THEORY

The ability of the ASA PS to predict adverse outcome frequency and severity in a nonlinear relationship can be explained by applying the Normal Accident Theory, a well-known system theory that relates the complexity of the system and interaction between components to the frequency and severity of system failures (ie, accidents).7,8 Acceptance of this theory relies on the establishment of 2 conditions: first, that anesthesia care is a component of a system; and second, that differences between ASA PS categories represent an increase in the interactive complexity between system components. Both of these premises depend on meeting descriptive criteria and not measurable (quantifiable) criteria.

The ability of the ASA PS to predict adverse outcome frequency and severity in a nonlinear relationship can be explained by applying the Normal Accident Theory, a well-known system theory that relates the complexity of the system and interaction between components to the frequency and severity of system failures (ie, accidents).

Observations on demographics from a large urban medical center patient population demonstrate the relationship between ASA PS categories and the relative frequencies of procedure type, age, use of inpatient postoperative care, and relative value units (RVU) as indirect and descriptive evidence that interactive complexity should increase with each increment in ASA PS category.

OBSERVATIONS

Frequency and Severity of Adverse Perioperative Outcomes

Data collected over a 10-year period in a single practice support the hypothesis that the frequency (Table 2) and severity (Table 3) of adverse perioperative outcomes, as previously defined in the literature,3,12 are related to the interactive complexity of system components. The correlation between ASA PS and adverse outcome rate and severity was very strongly positive. Application of a previously described structured peer-review model3,12 indicated that over 90% of these adverse outcomes were consistently due to system failure (Figure 1). Furthermore, our data confirm that the relationships between ASA PS and adverse outcome rate and severity are nonlinear (Figure 2). The magnitude of both the frequency and severity of adverse events continually increased with rising ASA PS rank order. The nonlinearity can be demonstrated by graphical analysis (Figure 2) and by the calculated odds ratios, which continually increased with ASA PS rank order (Table 1).

Figure 1.

The annual distributions of failure types (human or system) are shown. System failures annually accounted for over 90% of all adverse outcomes. Abbreviation: ASA PS, American Society of Anesthesiologists Physical Status.

Figure 2.

The average annual adverse outcome rate for all adverse outcomes by American Society of Anesthesiologists Physical Status (ASA PS) is shown with the annual variance for each ASA PS category given as standard deviation. ASA 5 shows the highest variance due to the relatively small numbers of ASA 5 patients per year. A trend line is plotted showing an exponential relationship between ASA PS rank order and average annual adverse outcome rate.

Table 2. Frequency of Adverse Perioperative Outcomes by American Society of Anesthesiologists Physical Status*.
 TotalHuman FailureSystem Failure
Total CasesTotal Adverse Outcomes%Odds RatioCI + 95%CI − 95%No.%No.%
  • *

    Frequency of adverse outcomes as a percentage of total cases for each American Society of Anesthesiologists Physical Status (ASA PS) group. The nonlinearity of the relationship between ASA PS and adverse perioperative outcomes can be seen in the odds ratio, which continually increased with each incremental increase in ASA PS.

ASA 149,3112160.441.00140.032020.41
ASA 2145,9779530.651.491.731.29460.039060.62
ASA 373,5399911.353.113.602.68520.079391.28
ASA 418,5893782.034.725.583.99250.133531.90
ASA 5980848.5721.3127.6516.4210.10838.47
Total288,3962,6220.911380.1024838.47
Table 3. Severity of Adverse Outcomes by American Society of Anesthesiologists Physical Status*.
 Outcome Severity ScoreTotal % 
12345Weighted Score
No.%No.%No.%No.%No.%Average 
  • *

    Distribution of outcome severity for adverse outcomes occurring in each American Society of Anesthesiologists Physical Status (ASA PS) group. Adverse outcomes were characterized as having 5 levels of severity, with level 1 being minor and level 5 being death. Level 1 and 2 adverse outcomes occurred largely among ASA 1 and 2 patients. More severe adverse outcomes (level 4 and 5) were predominantly seen in ASA 4 and 5 patients.

ASA 120.9016676.904621.3010.5010.50100.002.23
ASA 2171.8066069.3024125.30212.20131.40100.002.320.09
ASA 360.6063764.3024524.70585.90454.50100.002.490.17
ASA 482.1015039.7010628.00318.208322.00100.003.080.59
ASA 5 0.0056.0078.3022.407083.30100.004.631.55
Totals331.30161861.7064524.601134.302128.10100.002.56

Why should we apply the Normal Accident Theory to adverse perioperative outcomes? In an editorial in Anesthesiology written to accompany a paper by Owens, Arthur Keats opined that the ASA PS, despite its subjective nature, was a valuable clinical classification system that provided a link to outcomes.9,10 In his editorial, Keats went on to pose a provocative and paradoxical question: “Since Physical Status concerns only pre-operative ‘sickness’ and by definition does not include risks related to anesthesia and operation, why should any correlation (with outcome) exist at all?”10 There are 2 important factors to consider in answering this question: first, there is in fact a continuum of physical status from healthy to deathly ill represented incrementally by the 5 ASA PS categories in rank order; second, most of our adverse outcomes were identified as meeting our criteria for system failures.

“Since Physical Status concerns only pre-operative ‘sickness’ and by definition does not include risks related to anesthesia and operation, why should any correlation (with outcome) exist at all?”

ASA PS category assignment is based on an evaluation of a patient to determine if they fit an arbitrary definition that would place them in a group along a spectrum from healthy to deathly ill. This continuum is significant in that it also describes an indirect relationship between ASA PS and use of healthcare services. For example, there is a well-known relationship between age and healthcare use that can be measured in amount, cost, and intensity of healthcare services used.11 Demographic data from our practice indirectly show this relationship, which can be seen in the advancing age of the increasing ASA PS categories, the types of procedures associated with each ASA PS group, the relative work of the anesthesiologist (RVU and procedure length) and thus cost, and the increasing use of inpatient care as ASA PS increases (Tables 4 and 5). The increasing intensity of healthcare services associated with failing health (increasing ASA PS) brings with it additional complexity. This complexity is associated with the patient directly through additional healthcare providers, medications, tests, fragmented medical records, and disease progression. It is also associated with the healthcare systems caring for the patients by virtue of the added involvement with healthcare payers, both public and private, who impose regulations on providers and providing institutions. Over 90% of all adverse outcomes in our practice were considered through consensus of peer review to be system failures by definition.12 To understand system failures, it is important to understand the nature of the system that is failing and how it fails.

The increasing intensity of healthcare services associated with failing health (increasing ASA PS) brings with it additional complexity. This complexity is associated with the patient directly through additional healthcare providers, medications, tests, fragmented medical records, and disease progression.

Table 4. Patient Demographics and Procedure Characteristics of American Society of Anesthesiologists' Physical Status Groups*.
CasesASA 1ASA 2ASA 3ASA 4ASA 5Totals
  • Abbreviation: ASA, American Society of Anesthesiologists.

  • *

    Patient demographics and procedure characteristics for each American Society of Anesthesiologists Physical Status group. Procedure characteristics include relative value units, case length, and 10 most frequent procedures (Current Procedural Terminology codes).

No. (%)49,311 (17.1)145,977 (50.6)73,539 (25.5)18,589 (6.4)980 (.3)288,396
Admission status, % of total
 Ambulatory67.446.632.26.60.843.6
 Inpatient16.732.745.779.294.536.6
 Same Day Admission15.820.622.114.13.619.7
 Trauma0.20.10.10.21.10.1
Gender, % of total      
 Female56.069.856.748.041.062.6
 Male44.030.243.352.059.037.4
Age, % of total, years      
 <1836.214.87.14.44.315.8
 18–3941.532.69.84.86.926.4
 40–5918.827.125.823.726.525.1
 60–793.121.042.750.848.425.5
 80+0.44.514.716.213.97.2
Relative value units, % of total
 <6 Base Units87.875.168.340.534.872.0
 7–12 Base Units11.823.724.915.929.621.9
 13–18 Base Units0.41.23.96.118.02.3
 19+ Base Units0.00.12.937.617.63.9
Case length, anesthesia time in minutes, % of total,
 <9045.134.330.520.222.434.3
 90–18035.636.37.826.130.635.8
 180–36015.721.523.526.130.721.3
 360–5402.45.06.323.511.36.1
 5401.23.21.94.15.02.6
Total number of different procedures2324369630741262245 
Table 5. Frequency of Ten Most Common Procedures.
Top Ten ProceduresNumber of Cases (% of Total)
ASA 1ASA 2ASA 3ASA 4ASA 5
  1. Abbreviation: ASA, American Society of Anesthesiologists; CPT, Current Procedural Terminology.

1 CPT Code49,505 (3.4)1967 (8.8)66984 (4.6)33,512 (7.9)31,500 (9.2)
2 CPT Code42,820 (3.2)66,984 (4.0)11,042 (2.3)31,600 (4.1)49,000 (5.5)
3 CPT Code54,161 (2.4)59,514 (4.0)36,533 (2.1)33,511 (3.5)31,600 (4.1)
4 CPT Code19,120 (2.4)47,562 (2.3)36,821 (1.8)31,500 (3.4)31,603 (3.9)
5 CPT Code69,436 (2.3)59,409 (2.0)36,825 (1.8)33,405 (3.4)35,820 (3.5)
6 CPT Code47,562 (2.2)49,505 (1.8)36,800 (1.6)33,513 (3.2)39,000 (3.4)
7 CPT Code59,410 (2.1)19,120 (1.7)27,447 (1.4)33,430 (2.9)21,750 (2.8)
8 CPT Code44,950 (1.9)59,410 (1.6)27,236 (1.4)33,519 (2.7)33,975 (2.3)
9 CPT Code41,899 (1.9)58,150 (1.5)36,571 (1.2)33,510 (2.5)33,405 (2.2)
10 CPT Code58,670 (1.5)27,447 (1.4)35,556 (1.2)31,603 (2.1)35,081 (2.0)
Total487533 (25.3)449068 (23.3)351815 (18.3)329103 (17.1)306914 (15.9)

UNDERSTANDING SYSYTEMS

Systems are ubiquitous, diverse, and often have unclear boundaries; they can be organizations as well as processes. There are many different types of systems. Anesthesiologists work in a system that meets the definition of a transformative system. For the sake of discussion, we can call this system an anesthesia-requiring procedural system (ARPS). The patient-centered purpose of this system is to transform the patient from 1 state to another. That is to say, there is a desire on the part of the patient to be transformed from a state of sickness to health, from bad physiology to good physiology, from esthetically unappealing to esthetically appealing, from dying to living, or from a lack of knowing to knowing (diagnosis or prognosis). It is in short a purely transformative process. More accurately, it represents a family of transformative processes that are infinitely variable in terms of components (eg, different patients, different anesthesiologists, different surgeons, number of consultations, different locations, number and type of diagnostic tests, different devices, different drugs, different hospitals). The patient represents the raw material that is being transformed and thus is an integral part or component of the system. As a component, the complexity of the patient's medical problems increases the complexity of the system. Charles Perrow, a noted organizational theorist, wrote that “Irretrievably complex systems include systems that transform their raw materials (rather than fabricate).”7

Transformative Systems

The ARPS meets the definition of a transformative system. Although it is not uncommon to refer to this system in terms of productivity and efficiency, it is not involved with manufacturing. Similarly, despite analogies to piloting an airplane, the ARPS is not an operational system. The personnel working within the system are not operators of the system; they do not control the system or oversee the system, although they may appear to do so depending upon the context of what they are doing. Instead, anesthesia providers are actually components of this transformative system. With the anesthesiologist considered part of the ARPS, all adverse events that we previously defined as anesthesia-related human failure could be considered system failures. In other words, anesthetist error is merely a system component failure.

With the anesthesiologist considered part of the anesthesia-requiring procedural system, all adverse events that we previously defined as anesthesia-related human failure could be considered system failures. In other words, anesthetist error is merely a system component failure.

System Components and Interactive Complexity

Systems have in common the fact that they are made up of parts, the smallest components that become grouped together into units that are functionally related. Subsystems are an array of units, and functionally related subsystems form systems. Systems can be characterized by their complexity; however, it is important to clarify that complexity is not the opposite of simplicity in this context. Instead, systems are considered linear or complex, that is to say either linear or nonlinear (also called complex). This classification depends on the nature of the interaction of the systems' components. Complex interactions result in more than 1 interaction between components; linear interactions are single interactions between components. Complex system component interactions increase exponentially, whereas in contrast linear systems do not. Complex systems are characterized by having components that are in close proximity, with many feedback loops with multiple interacting controls, and with indirect information and limited understanding of the system. Such systems often have many interconnected subsystems. Good examples of complex systems include nuclear power plants or global weather systems. Linear systems, on the other hand, are characterized by having components that may be spatially segregated with few feedback loops. When linear systems occur they are single purpose, with segregated controls receiving direct information, and there is extensive understanding of the system. A typical assembly line in a manufacturing plant is likely to be more linear than complex. Most systems are a mixture of both complex and linear systems, and the admixture of both system types determines the nature of the system's overall complexity.

Coupling

Another property of systems is the tightness of coupling between components within the system. Tightly coupled systems are those where the process cannot go on standby (time-dependent coupling), making it difficult to stop the system and fix something. Tightly coupled systems have sequences that are largely invariant, so that it is not possible to proceed without completing the prior sequence. This leads to systems where the overall design is largely invariant such that there is only 1 way to proceed. Such tightly coupled systems are unable to tolerate slack so that quantities must be precise and resources cannot easily be substituted. Quality has very little room to vary without causing a tightly coupled system to fail.

Interactive complexity between components may depend on the components themselves. Multimodal components are those components that have the capacity to perform more than 1 function depending on circumstances, and are by their very nature more complex than a component that performs only 1 function. Multimodal components are a feature of more complex systems and are frequently part of feedback systems. An anesthesiologist as part of an ARPS would be a good example of a very complex multimodal component or subsystem.

Thus, the ARPS has all of the characteristics of interactive complexity with tight coupling between components. The components of an ARPS consist of both humans and machines. Anesthesiologists, surgeons, consultants, nurses, operating room support staff are all examples of components representing complex subsystems. Mechanical components include anesthesia machines, monitors, electrical systems, surgical devices, and instruments. Other systems interact as well, including hospital systems and hospital subsystems, regulatory systems, and payment systems. Many of these systems can be characterized by having many feedback loops with multiple interacting controls. There is often only indirect information that is used to modulate the transformative process, and there can be little doubt that our understanding of this transformative process is limited, thus the need for continued research. For example, the measurement of blood pressure is part of a feedback loop between the anesthesiologist and the patient. The information obtained provides indirect information on the patient's condition that the anesthesiologist uses to make decisions and carry out action. There is also tight coupling between the steps that are largely invariant. For example, it is not possible to stop the process and fix something because there are time constraints due to physiological limitations. The action of the surgery and the action of anesthesia are very tightly coupled and not easily separated.

Given the variety of patients and procedures, there is a variety of interactive complexity and tightness of coupling. For example, an inguinal hernia repair done as an outpatient procedure in a healthy (ASA 1) man requires a much less complex ARPS in terms of component interactions, than a sicker (ASA 4) patient having a cardiac valve replacement. However, an ASA 4 patient having the same outpatient hernia repair could easily require added monitoring, more testing, and more care postoperatively than the healthy patient. The ARPS complexity that a patient will be exposed to during the transformative process is roughly represented by the ASA PS category. ASA PS arbitrarily groups the patient along the continuum of healthy to deathly ill and thus to the level of complexity required by their care. Increasing complexity, in a tightly coupled system, leads to an exponential increase in component interaction, and subsequently an exponential increase in the risk of system failure. In contrast, linear systems having an increase in complexity will not exhibit exponential increases in the likelihood or severity of failure. We have demonstrated that increasingly complex transformative ARPS systems, as represented by increasing ASA PS, behave with an exponential increase in the rate of associated system failures and increasing severity of those failures.

INCIDENTS AND ACCIDENTS

When things go wrong with systems, they go wrong at different levels. Incidents are generally adverse outcomes that are relatively minor and are those that are easily reversed or recoverable. Incidents occur when there is a part failure (smallest system component) or failure at the unit level (functionally related parts). Accidents are more severe and occur when subsystems (arrays of units) or systems fail.13 Our adverse outcomes were characterized as having 5 levels of severity, with level 1 being minor and level 5 being death. Level 1 and 2 adverse outcomes occurred largely among ASA 1 and 2 patients. These adverse outcomes were extremely minor and can be considered incidents where there was recovery or reversibility. More severe adverse outcomes (level 4 and 5) were predominantly seen in ASA 4 and 5 patients (Table 2). These outcomes were severe irreversible accidents.

A number of years ago, Gaba et al. developed an argument by analogy that offered a rational explanation for adverse outcomes using the concepts developed by Dr. Charles Perrow, whose book Normal Accidents details his analysis of events leading to major catastrophes including the Three Mile Island accident.7,8 Perrow theorized that accidents occur as a normal part of systems despite fail-safe mechanisms and incentives to prevent failures. Perrow terms this phenomenon normal accidents. In essence these are system failures, and he considers them inevitable. Gaba et al. argued that anesthesiologists are operators of complex systems to which Perrow's principles apply. Our observations force a reconsideration of the scope and characteristics of the system, and therefore the scope of what is an inevitable catastrophe or normal accident. Although our observations support the argument that Gaba et al. proposed, we would expand the concept of the system to include the administration of anesthesia, and anesthesiologists themselves, as a subsystem component of a much broader transformative system.

Perrow theorized that accidents occur as a normal part of systems despite fail-safe mechanisms and incentives to prevent failures. Perrow terms this phenomenon normal accidents. In essence these are system failures, and he considers them inevitable.

Transformative systems are frequently closely monitored and managed by use of feedback loops. Many believe that more monitoring can lead to greater control and more safety. However, it can also lead to more interactive complexity with unintended consequences. The idea of identification of a critical incident is an axiomatic concept and generally accepted in the patient safety literature.14–16 The axiom is if a critical incident can be identified early, then it should be possible to recover from that incident and prevent an adverse outcome from occurring. Cooper et al. identified and stressed this concept as fundamental to monitoring and to anesthesia safety.14–16 Philosophically, this implies that an anesthesiologist is in control of the anesthetic care of the patient and should be able to modify outcome by skilled and knowledgeable action. This anthropocentric concept emphasizes the importance of human errors and of individual responsibility, an idea that supports certain strategies for quality improvement and medical safety, including designing systems that minimize human errors.17 Moreover, this idea is fundamental to the rationale for monitoring; namely, it is imperative to use monitors to identify critical incidents as early as possible and then be able to take timely and appropriate action before something adverse occurs. However, paradoxically, if the anesthesiologist is a component of the system and has the capacity to take an action (or not take action) that is ineffective or wrong, then identification of preincident factors can also result in a system failure and an adverse outcome depending on the nature of the components (the anesthesiologists and the monitor for example). An anesthesiologist, viewed as a common-mode multimodal component of a transformative system, may also react to information in a perfectly rational and appropriate way that leads to consequences that might not ordinarily be expected. Because of tight coupling, recovery is frequently not possible and may thus lead to an accident.

If system failures are inevitable, efforts to alter outcomes by introducing system changes can have unintended consequences, because added complexity could result in more, rather than fewer, unintended problems. These inevitable normal accidents are exemplified by use of pulse oximetry and pulmonary artery catheters as important monitoring devices that would give more information and more opportunity to limit problems through feedback control.18 A large, randomized, controlled trial of pulse oximetry was conducted in Denmark between February 1989 and June 1990.19,20 During this time, 20,802 patients were randomly assigned to be monitored with or without the use of pulse oximetry in the operating room and postanesthesia care unit (PACU). In general, demographic data, patient factors, and anesthetic agents were distributed evenly between the 2 groups. Despite a 19-fold increase in the incidence of diagnosed hypoxemia in patients monitored by oximetry (P < 0.00001), the investigators could not demonstrate a significant benefit in terms of outcome. However, use of this monitor led to an increase in adverse events such as longer PACU stays, more intensive care unit admissions, increased use of supplemental oxygen, and more frequent administration of naloxone. Likewise, a prospective-cohort study also showed an increase in associated adverse outcomes such as increased hospital stay and 30-day mortality in patients monitored with pulmonary artery catheters.21 Both of these monitors add to system complexity and component interaction, and could be predicted to be associated with increased system failure with the potential for unintended consequences.

If it is assumed that the anesthesiologist is part of a large system, either as a component or a subsystem, then we are required to reexamine the central importance of the idea of individual control of, and recovery from, critical incidents. In short, the assumption shifts the point of view from an anthropocentric concept to one that is not, where the anesthesiologist becomes a system component and not a driver/controller of a system. Under this concept a human error becomes only 1 way in which this component can fail and diminishes the importance of the axiom of critical incident recovery in the performance of the APRS.

NORMAL ACCIDENT THEORY AND AMERICAN SOCIETY OF ANESTHESIOLOGISTS PHYSICAL STATUS

To explain the relationship between ASA PS and adverse outcome reported here, it is understood that ASA PS nominally identifies 5 groups that arbitrarily divide patients into increasing levels of failing health status. Each level of failing health status commits the patient undergoing a procedure that requires anesthesia to a unique transformative periprocedural system whose level of interactive complexity relates directly to that ASA PS. Thus, the patient's health links the patient to a level of ARPS complexity. The level of ARPS complexity is a determinate, and thus a predictor, of the chance of an adverse outcome or system failure. Therefore, the likelihood and severity of an adverse outcome is predominantly a normal and inevitable consequence of failures associated with each unique transformative system. Applying this theory, one predicts that increasing complexity associated with increasing ASA PS should be accompanied by increasing likelihood of an adverse outcome and increasingly severe outcomes. Normal Accident Theory explains our observations and provides a mechanism for the occurrence of the vast majority of the adverse outcomes we observed, but that is not proof of its validity. Because of the difficulty of designing experiments to prove this theory in this context, it is likely that confidence in its validity will come slowly from a continual capacity to predict and explain.

Each level of failing health status commits the patient undergoing a procedure that requires anesthesia to a unique transformative periprocedural system whose level of interactive complexity relates directly to that ASA PS.

There is little doubt that the specialty of anesthesiology has made significant advances since 1941, yet the perioperative mortality rate (death within 48 hours of surgery) has not changed much in over 30 years, and neither has the anesthesia-related mortality rate changed significantly.12 Still, it is common knowledge that we are anesthetizing and operating on much sicker patients and vastly expanding the opportunity for patients to take advantage of promising perioperative technologies. Perhaps as anesthesiology advances knowledge and technology and extends care to an ever-broader spectrum of patients, we eventually increase our system complexity and inevitable failure rate to a point where we are achieving the same limits of failure that our society was willing to tolerate in the past.

CONCLUSION

There seems to be an enduring relationship between the ASA PS and adverse perioperative outcome, not only in anesthesia, but also in surgery.5 It is clear that ASA PS has the capacity to predict adverse outcome rate and the severity of those outcomes in a defined practice. These relationships are nonlinear and appear to be exponentially related to the rank order of the ASA PS category. The vast majority of these failures are judged to be system failures by peer review. Analysis also shows that the anesthesia requiring procedural system meets the definitional criteria for a transformative system and has all of the features of a system that is irretrievably complex with many tightly coupled components. Failures in such a system are inevitable. The general mechanism of failures, the likelihood of failure, and the severity of failure can all be explained by Normal Accident Theory. As a sociologist, Perrow realized that the development of more complex systems created a situation with increasing likelihood of catastrophic failures that are a normal part of these systems. Perrow's work raises several important questions. We need to ask how to reduce the likelihood of system failures, but also what are the implications for a society trying to learn to live with the technologically advanced systems that bring many positive benefits while creating new problems (ie, normal accidents). Similarly, our specialty needs to ask these questions about anesthesia requiring procedural systems. Unfortunately, without a means of measuring interactive complexity between components there is no easy way to prove this theory as to why ASA PS predicts both the rate and severity of adverse outcomes in any rigorous sense; rather its acceptance, like the theory of evolution, will either come with a consistent ability to explain and predict future observations or it will fail and ultimately be rejected.

Acknowledgements

The authors thank Charles Perrow, PhD, Professor Emeritus of Sociology, Yale University, for his review of the manuscript and insightful comments.

DISCLOSURES

Potential conflict of interest: Nothing to report.

Ancillary