The differential role of basic science knowledge (e.g., physiology and biochemistry) in solving problems of varying complexity and the differences between subjects at different levels of expertise (Patel et al., 1994) have been a source of controversy in the study of medical cognition (Patel & Kaufman, 1 995) as well as in medical education and AI. As expertise develops, the disease knowledge of a clinician becomes more dependent on clinical experience, and clinical problem solving is increasingly guided by the use of exemplars and analogy and becomes less dependent on a functional understanding of the system in question. However, an in-depth conceptual understanding of basic science plays a central role in reasoning about complex problems and is also important in generating explanations and justifications for decisions.
Researchers in AI were confronted with similar problems in extending the utility of systems beyond their immediate knowledge base. Biomedical knowledge can serve different functional roles depending on the goals of the system. Most models of diagnostic reasoning in medicine can be characterized as being shallow. A shallow medical expert system (e.g., MYCIN and INTERNIST) reasons by relating observations to intermediate hypotheses that partition the problem space and further by associating intermediate hypotheses with diagnostic hypotheses. This is consistent with the way physicians appear to reason.
There are other medical reasoning system models that propose a "deep" mode of reasoning as a main mechanism. Chan-drasekeran, Smith, & Sticklen, (1989) characterize a deep system as one that embodies a causal mental model of bodily function and malfunction, similar to the models used in qualitative physics (Bobrow, 1985). Systems such as MDX-2 (Chandrasekeran et al., 1989) or QSIM (Kuipers, 1987) have explicit representations of structural components and their relations, the functions of these components (in essence their purpose), and their relationship to behavioral states.
To become licensed physicians, medical trainees undergo a lengthy training process that entails the learning of biomedical sciences, including biochemistry, physiology, anatomy, and others. The apparent contradiction between this type of training and the absence of deep biomedical knowledge during expert medical reasoning has been pointed out. To account for such apparent inconsistency, Boshuizen and Schmidt (1 992) proposed a learning mechanism -knowledge encapsulation. Knowledge encapsulation is a learning process that involves the subsumption of biomedical propositions and their interrelations in associative clusters under a small number of higher-level clinical propositions with the same explanatory power. Through exposure to clinical training, biomedical knowledge presumably becomes integrated with clinical knowledge. Biomedical knowledge can be "unpacked" when needed but is not used as a first line of explanation.
Boshuizen and Schmidt (1992) cite a wide range of clinical reasoning and recall studies that support this kind of learning process. Of particular importance is the well-documented finding that with increasing levels of expertise, physicians produce explanations at higher levels of generality, using fewer and fewer biomedical concepts while producing consistently accurate re sponses. The intermediate effect can also be accounted for as a stage in the encapsulation process in which a trainee's network of knowledge has not yet become sufficiently differentiated, resulting in more extensive processing of information.
Knowledge encapsulation provides an appealing account of a range of developmental phenomena in the course of acquiring medical expertise. The integration of basic science in clinical knowledge is a rather complex process, however, and encapsulation is likely to be only part of the knowledge development process. Basic science knowledge plays a different role in different clinical domains. For example, clinical expertise in perceptual domains, such as dermatology and radiology, necessitates a relatively robust model of anatomical structures that is the primary source of knowledge for diagnostic classification. In other domains, such as cardiology and endocrinology, basic science knowledge has a more distant relationship with clinical knowledge. The misconceptions evident in physicians' biomedical explanations would argue against their having well-developed encapsulated knowledge structures in which basic science knowledge can easily be retrieved and applied when necessary.
The results ofresearch into medical problem solving are consistent with the idea that clinical medicine and biomedical sciences constitute two distinct and not completely compatible worlds with distinct modes of reasoning and quite different ways of structuring knowledge (Patel, Arocha, & Kaufman, 1994). Clinical knowledge is based on a complex taxonomy that relates disease symptoms to underlying pathology. In contrast, biomedical sciences are based on general principles defining chains of causal mechanisms. Learning to explain how a set of symptoms is consistent with a diagnosis therefore may be very different from learning how to explain what causes a disease. (See Buehner & Cheng, Chap. 7, for a discussion of causal learning.)
The notion of the progression of mental models (White & Frederiksen, 1 990) has been used as an alternative framework for characterizing the development of conceptual understanding in biomedical contexts. Mental models are dynamic knowledge structures composed to make sense of experience and to reason across spatial or temporal dimensions (see Johnson-Laird, Chap. 9). An individual's mental models provide predictive and explanatory capabilities of the function of a given system. The authors employed the progression of mental models to explain the process of understanding increasingly sophisticated electrical circuits. This notion can be used to account for differences between novices and experts in understanding circulatory physiology, describing misconceptions (Patel, Arocha, & Kaufman, 1994), and explaining the generation of spontaneous analogies in causal reasoning.
Running a mental model is a potentially powerful form of reasoning but it is also cognitively demanding. It may require an extended chain of reasoning and the use of complex representations. It is apparent that skilled individuals learn to circumvent long chains of reasoning and chunk or compile knowledge across intermediate states of inference (Chandrasekaran, 1 994; Newell, 1990). This results in shorter, more direct, inferences that are stored in long-term memory and are directly available to be retrieved in the appropriate contexts. Chandrasekaran (1 994) refers to this sort of knowledge as compiled causal knowledge. This term refers to knowledge of causal expectations that people compile directly from experience and partly by chunking results from previous problem-solving endeavors. The goals of the individual and the demands of recurring situations largely determine which pieces of knowledge get stored and used. When physicians are confronted with a similar situation, they can employ this compiled knowledge in an efficient and effective manner. The development of compiled knowledge is an integral part of the acquisition of expertise.
The idea of compiling declarative knowledge bears a certain resemblance to the idea of knowledge encapsulation, but the claim differs in two important senses. The process of compiling knowledge is not one of subsumption or abstraction, and the original knowledge (uncompiled mental model) may no longer be available in a similar form (Kuipers & Kassirer, 1984). The second difference is that mental models are composed dynamically out of constituent pieces of knowledge rather than prestored unitary structures. The use of mental models is somewhat opportunistic and the learning process is less predictable. The compilation process can work in reverse as well. That is to say, discrete cause-and-effect relationships can be integrated into a mental model as a student reasons about complex physiological processes.
Was this article helpful?