Thick Versus Thin: Description Versus Classification in Learning From Case Reviews

SEE RELATED ARTICLE, P. 251.

[Ann Emerg Med. 2008;51:262-264.]

Morbidity and mortality (M & M) conferences have a prominent place in medical education. In addition, they are social accountings, attempts to align discrepant outcomes with expectations1; as such, they are an important part of the legitimatizing self-regulation of a profession. Such conferences and other types of quality review are ubiquitous in modern health care; 94% of accredited ED residencies hold regular M & M conferences,2 and more than 90% of hospitals report regular mortality reviews.3 However, given the ubiquity and importance of these reviews, it is surprising how little about them appears in our literature; a PubMed search (“morbidity OR mortality” in title or abstract AND “education OR quality OR peer review” in any field) of 4 prominent North American emergency medicine journals produced only 5 references that might be considered at least marginally relevant to these reviews.2, 3, 4, 5, 6

The report in this issue of Annals by Cosby et al7 shines a welcome light into this dark space. Like most forays into underexplored areas, its value derives in somewhat unexpected ways. Cosby et al7 analyzed 636 cases referred for review during a 15-year period; rather than relying simply on the chart, they attempted to compile “thick descriptions” of the cases from contemporaneous interviews of those involved, notes of the subsequent discussions in multiple forums, the medical record itself, and any formal recommendations that resulted. Based on this rich content, they assigned problems to categories roughly representing phases of care and attempted to identify classes of factors contributing to the problem. They wisely did not make these classifications mutually exclusive, but rather viewed them as sets of descriptive attributes, thus avoiding the pitfall of an overly rigid standardization that would lead to distortions and workarounds. The development of such classifications is not a simple problem, and the authors’ thinking obviously and appropriately evolved over time due to prolonged contact with their data.

Attempts at classification are natural responses to unruly, heterogeneous, “messy” phenomena. One might think that constructing classifications would be a relatively straightforward task, and perhaps it would be, if only classifications were “things in the world” instead of “things in our heads.” Borges8 provides a famous example of the difficulties of classification in his account of a certain Chinese encyclopedia, the Celestial Emporium of Benevolent Knowledge, which divided animals into “those belonging to the emperor; embalmed ones; those that are trained; suckling pigs; mermaids; fabulous ones; stray dogs; those included in the present classification; those that tremble as if they were mad; innumerable ones; those drawn with a very fine camelhair brush; others; those that have just broken a flower vase; and those that from a long way off look like flies.”

Lest we be too critical of the Celestial Emporium, we should recall that early mortality tables listed causes of death in such classes as “fainted in a bath,” “frighted,” “itch,” and “suddenly.”9 These classifications have changed over time, although it is unlikely that ultimately fatal pathologies have changed nearly as much; the history of the International Classification of Diseases (ICD) system shows that about a century ago, it was possible to die from “senility” or being “worn out”; today, old age is hardly acceptable as a cause of death.9 One wonders in what regard the 10th revision of the ICD will be held a century hence.

A classification scheme is fundamentally an ontology—an answer to the question, What sorts of things are there?—and so depends as much on understandings of how the world works, who does the classifying, and for what purpose as it does on the nature of things in the world. Each possible classification validates one point of view and silences others.10 Because of this, classifications often tell us more about the classifiers than the classified. In this study, the choice of cases for review and the selection of factors deemed important say as much or more about the reviewers than the reviewed; they provide insight not into the types of things that can go wrong in emergency care, but rather the types of things that we think we ought to be able to do something about. For example, the clinician-reviewers in this study tended to find problems in diagnostic decisionmaking and technical skills, areas to which they dedicate great effort in mastering and teaching. They were less likely to note factors such as vague or conflicting goals; still other factors, such as multitasking or time pressure, may have been seen as immutable givens and thus no more useful as explanations than gravity in explaining airplane crashes. The study is no less valuable because of this, as long as we realize that the meaning it conveys differs from a nominal reading of its results.

Because of the relative recency of the study of safety in clinical work and the length of time spanned by the cases studied, the authors’ classification scheme is eclectic, pieced together from disparate sources; consequently, it tends to mix dimensions and definitions in a manner reminiscent of the Chinese encyclopedist’s list. For example, events classified as communication failures might also reasonably have been called cognitive failures or even system failures (for example, if ambient noise levels were high). None of these views is inherently right or wrong, but they lead to different understandings that can be more or less useful, depending on purpose and context. The development of these taxonomies of “errors” has obsessed much of the patient safety field for some time, but the paucity of useful results from these efforts calls their value into question. Their categories inevitably become fuzzy and overlapping; rather than illuminating the complexity of accidents and incidents, they tend to obscure it by averaging out complex internal detail.11

However, the value of the Cosby et al7 work lies less in the results than in the process: the thick descriptions rather than the thin classifications. Descriptive analyses of accidents and incidents are often superficial in health care, at least in comparison to that in other hazardous industries. The authors’ methods of developing their thick descriptions—the rich, nuanced details of the complex work involved in emergency care, which sometimes leads to accidents and incidents—come close to approximating the procedures described more formally in the accident investigation literature.12, 13, 14, 15 It is from processes like these—detailed explications of individual cases, deeply situated in complex contexts—that insights leading to useful reductions in hazards are likely to emerge. Enhancing this analytic process promises to be more useful than further refinement of ever more elaborate taxonomies of “errors.”

The Cosby et al7 approach should be emulated and could be enhanced 2 ways. First, a greater attempt at ensuring “requisite variety” on the analytic team is important. If the team is provided a sufficiently diverse set of backgrounds, viewpoints, skills, and interests, then hidden assumptions are exposed; a broader repertoire of options, tactics, and tools is made available; tacit knowledge is made more explicit; and more interpretations and preferences are expressed.16, 17, 18, 19 Safety in health care has become overly “medicalized,”20 leading to a narrowing of what can be seen and a consequent narrowing of what can be done: broadening the skill set of investigative teams will be critical to further progress. This broadening should go beyond including different clinical specialties and professions to also include psychologists or engineers because clinical failures are more problems of psychology and engineering than they are of medicine.21

Second, developing specific skills in the elicitation or reconstruction of events in their context would usefully supplement simpler recountings of “what happened.”12, 13, 14, 15 Individuals involved in adverse events necessarily try to make sense of them and in this process may consciously or unconsciously simplify, linearize, or otherwise distort their recall based on knowledge of the result. Thus, recountings are never simply “replays” of what happened, but rather reconstructions that incorporate the participants’ preexisting understandings and attempts to add meaning.22 These reconstructions are often socially mediated, and this mediation can either add or remove value, depending on the skill of the elicitors. Skilled reconstructions of context, especially when performed by analysts who are not in some way stakeholders, would considerably enhance understandings of hazards and how they are dynamically, unconsciously, and almost invisibly negotiated by clinical workers23 in a manner generally so skillful that what could have been a disaster is usually no more than a minor perturbation in the smooth flow of high-tempo operations24 and only rarely becomes visible in the aftermath of an adverse event or a “near miss.”

A final lesson from the Cosby et al7 article is wise advice to clinician-researchers: “Write about what you do.” Social structures, such as M & M conferences, do not persist unless they benefit someone, and very common structures are often taken to be normal parts of the world, not worthy of study; but if an activity is important enough for clinicians and educators to willingly invest their time and effort, it is important enough to study. It is through understanding what might be regarded as mundane by a group that we can gain insight into its characteristic systems of thought.25 Such undertakings are risky because they move into uncharted territory, requiring novel and unfamiliar methods and perhaps leading to confusion before they produce insight; studying buffered versus unbuffered lidocaine would be safer and can be done oh so rigorously. But it will hardly matter, whereas explorations of the previously unexplored might.