Rationalizing Medical Work: Decision-Support Techniques and Medical Practices

Marc Berg undertakes in this very interesting volume to analyze the application of decision-support tools—computer-based support systems, protocols, and clinical decision analysis—to medical practice. In doing so he raises several interesting questions about the nature of medical diagnostic and therapeutic practice and about the use of computer-based and other “rationalizing” technologies in contemporary work and culture.

Berg begins with an incisive discussion of the post-World War II medical literature, tracing how arguments for the use of decision-support tools were developed, supported, and critiqued. His demonstration that each tool was shaped by a particular concept of what “science-based medicine” is, or should be, is well reasoned and provocative, and prompts the reader to reexamine what we understand by scientific practice.

The central section of the book considers the construction and implementation of several decision-support tools: a triage aid to help nurses decide which chest pain patients should be admitted to the coronary care unit, a computer program for the diagnosis of abdominal pain, a research protocol for an experimental breast cancer treatment. Based on these examples, Berg argues that structured, rationalized aids are usually transformed in practice by a variety of logistical, organizational, and value-based factors—staff skills and hierarchies, variations in patient presentation financial criteria, ethical questions, among many others; that where decision-support tools succeed, it is after the practice site is “disciplined to a formalism” (p. 79), that is, where structures are introduced or reinforced to support the tool’s operation; that this mutual transformation of tool and practice site by each other is highly localized in space and scope—and must be replicated and readjusted if the tool is introduced in another setting; finally, that even where the tool is successfully adapted for regular use, doctors and other staff routinely support, adjust, and retransform its operation. The tool and the practice in this way “co-produce” each other to form a new adaptive hybrid. In Berg’s view, both advocates and critics, wedded to prescriptive views of how decision-support tools “are supposed to work” or not to work, fail to understand the nature and potential of the way they actually do work.

Berg’s argument is intelligent and well researched through interviews, practice observations, and reading of the literature. It is probably my historian’s bias, but I think the book might have been stronger if he had placed more emphasis on his unique material and slightly less on the theoretical discussion, which, as he notes, complements and extends analyses of scientific practice and work by contemporary anthropologists and other scholars—Bruno Latour, Donna Haraway, Susan Leigh Star. His many interesting statements about the flow of medical practice, its embedded assumptions, and the work done to incorporate tools are supported with concise observations drawn rather disjointedly from his several cases, supplemented by only a few complete narratives showing us the internal processes. Particularly after the author tells us that very few tools have been introduced in more than a single site, or remained in permanent use, the reader is forced to question whether his model has reality only in localized instances.

One tool that might have received more intensive analysis is the use of the problem-oriented medical record introduced by Lawrence Weed, which has been almost universally instituted, at least in the United States, and which Berg discusses in his review of the literature, but to which he fails to return.

One of the principal technologies examined is the breast cancer research protocol. As Berg points out, research protocols are widely used and highly structured. But they also present a different rationale from decision tools introduced into established clinical practice. A research protocol carries with it the implication, often the promise, that existing rules and routines will change. This implicit assumption does not invalidate Berg’s model of “co-production,” but its operation will be different than in the other examples...