MEDICAL EXPERTISE:

Many questions, few answers

 

Cîmpian Erika Ildikó

Department of Psychology, “Babeş-Bolyai” University

cerka@coltrans.org.soroscj.ro

 

 

 

 

From a psychological point of view, establishing the diagnosis may be described as an ill-defined problem (Pople, 1982). The main goal is to match a sequence of information (concerning the patients’ physical condition) with a certain diagnostic category (Kerbeshian, 1991). The physician’s task is ill-defined at all levels.

In the process of medical problem solving, the assessment of the initial state and the construction of an adequate representation are limited by the fact that the data are presented in both incomplete and unstructured manner. The patient enumerates only a part of the information-package necessary to establish the diagnosis. These input-level data are undifferentiated, too. The patient could fail in giving some key-information for generation or elimination of certain hypothesis, because he can not make the correct distinction between relevant and irrelevant information. In the initial phase of clinical problem solving, the deficiencies are doubled by the impossibility to determine which information is missing or which of the unknown information will prove to be relevant. The process of representation construction goes on until the diagnosis is determined, or even beyond. Similarly, the operators, specifically the implementation of the operators, are ill-defined in the sense that it is unclarified on which information they have to be applied. Obviously, the relevance of an information is determined by the diagnostic context. For example, the same quantity of serum creatinine could be interpreted differently, depending on weight, protein intake, and pregnancy (Norman & al., 1992). It is difficult to circumscribe the search in the problem space – both biomedical and clinical knowledge will act as constraints in two directions: a.) to allow the generation of the most likely hypothesis; and b.) to make possible the elimination of invalid hypothesis (Kerbeshian, 1991). The final state – the correct diagnostic category - is also insufficiently defined. Moreover, the success is by no means certain, because the accuracy of the diagnosis is based on the reaction of the patient to the treatment, this way it may be only partially estimated. Usually, we assume that if the proposed treatment works, then the diagnosis was correct, yet the diagnosis remains “the best fitting hypothesis out of a number of possible candidates rather than the only possible answer” (Gilhooly & al., 1997, p.: 200). In order to make it more palpable how difficult this kind of task may be, let us remind ourselves that an internal medicine textbook includes approximately 220,000 data, as well as a cardiology textbook presents roughly 90,000 data (Mérő, 1997).

No doubt that medical expertise is one of the most intriguing topic of psychological research considering that there are major inconsistencies between findings in this domain and results of other special domains. In a variety of tasks, contrary to other examined domains, researchers (like Claessen and Boshuizen, 1985) failed to find any significant differences between experts and beginners, except the volume and organization of their knowledge base (Patel and Groen, 1986). But, as the study of medical cognition switched over to investigate the domain-specific knowledge instead of the general form of clinical reasoning, our knowledge concerning medical problem solving became more articulated.

During decades of research there was only a small number of trials of summarizing the results coming from different areas of expertise, searching for common mechanisms across domains. The shortcomings of theories [like Chase and Simon’s perceptual-structures theory (1973) or the later skilled-memory theory developed by Chase and Ericsson (1981)] results from the fact that they were developed based on observations gathered from a single domain tough they claimed that the conclusions could be generalized. They explained expert performance in terms of complex knowledge structures build up through the process of mastering, and retrieved from LTM in order to accomplish the domain-specific tasks. More recently Vicente and Wang (1998) attempted to set up a product theory as opposed to the above-mentioned process theories, revising many experiments from a series of domains, and tried to reinterpret the findings according to their basic assumptions. Before proceeding with the short presentation of the ecological theory, let emphasis that this theory do not question the role of the knowledge structures, instead try to account for phenomena insufficiently explained so far.

Vicente and Wang’s constraint attunement hypothesis set up from the idea that expertise can be view as “maximal adaptation to task constraints”(as Ericsson and Lehmann called it, 1996). Every task or domain has its own constraints; these are always determined as a function of the main goal. We will take into account different constraints depending if our purpose is to select a certain kind of information or it is to memorize the material. In a certain domain, experts have to deal with a set of specific tasks sharing usually the same goal- and constraints structure. These goal-relevant constraints, defined as relationships pertinent to the domain, will shape the way experts structure the stimuli. Adaptation or more precisely attunement to a specific problem environment allows introducing some order in complexity. Violation or elimination of all or a part of these constraints can seriously affect the information processing, and so, consequently diminishes the overall performance.

The purpose of this study is not to defend either the earlier, or the later theories; it would be more profitable to use both of them in order to develop a more complete picture of medical expertise.

The influence of expertise in processing medical cases

 

The main criterion for medical expert performance is high diagnostic accuracy, so we can not accept as a satisfactory measure for the ability to establish the correct diagnosis such an one-sided index like recall achievement of presented materials. Paraphrasing  Gobet’s words (1994, p.: 45) about chess players, physicians main occupation is not to recall experiments, but to make accurate diagnoses. As Vicente and Wang (1998) pointed out, we have to make a clear distinction between intrinsic tasks and contrived tasks. The former ones are “definitive features of that domain of expertise” (p.: 34), that is, problems that experts usually solve; the later ones are just parts of that domain and have an important roles in eliciting experts’ knowledge, but includes problems that experts usually do not encounter. Still, I shall review some of those recall- and recognition-experiments, which suggest not only better memory skills, but also essential differences in processing.

            Of course, experienced physicians recalled a great amount of presented data, twice as much as did inexperienced subjects, but their apparently indisputable advantage was observed only in conditions where non-directive, so called “incidental-memory” instructions were given. With explicit memorizing instructions, students have no problem catching up with their masters (Norman, Brooks, and Allen, 1989). Moreover, both experts and novices recalled critical and abnormal data better than non-critical and normal data regardless of instructions. Yet, some major differences in processing are suggested by the fact the experts recalled more non-critical data under incidental instructions then students, although in this condition their goal was simply to find a diagnosis instead of memorizing the data as in the intentional condition. It appears that once they identified the major disorder based on the critical information selected from the presented data, they allocated their cognitive resources in order to verify if there is any signs indicating some unusual or alternative explanations. Inexperienced subjects ignored non-critical data because they were searching for the explanations of the abnormal findings.

            Myles-Worsley, Johnston, and Simons (1988) reported similar results. Experts’ memory for abnormal X-ray films was as good as the memory for faces. Still, regarding the recognition of normal X-ray films there were at a lowest level then residents, or even novices. This unusual behavior indicates differences in processing, more precisely strong preferences for selectively process clinically relevant abnormal features over irrelevant variations of normal ones. In time they have acquired an elaborate schemata of normal features, which helps them to quickly check out the presence of normal parts, so that the processing resources remain available to detect distinguishing characteristics. The schemata guide the selection of relevant information and strike out irrelevant information, and in this case the abnormal features definitely are the relevant ones. Thus is understandable why is unacceptable to compare radiologists’ performance at normal versus abnormal X-ray films with chess players’ performance at real versus randomized or unfamiliar chess position. For radiologists both normal and abnormal films are meaningful and equally familiar. But their main goal is to detect any existing abnormality, thus is not surprising that they perform better in recognizing or recalling abnormalities. Each detected abnormality has its own sense, there is a natural underlying mechanism which lead to its development, as in chess every meaningful position is a part a sequence of regular moves, one step of a game that effectively could take place. For them the “meaningful board configuration” is in fact an abnormal film. Experts use both kind of knowledge (about normal, as well as about abnormal characteristics), but for different purposes; while elaborate knowledge of abnormal features is required to perform the diagnostic task, elaborate knowledge of normal films is necessary to allow this. According to the constraint attunement hypothesis presented by Vicente and Wang (1998), experts in radiology are more attuned to the goal-relevant information, that is, abnormalities, whereas novices (at this level of expertise) are more attuned to the irrelevant characteristics, which in this case are the normal features.

            Applying the same constraint attunement hypothesis, the authors above-mentioned also give a reasonable explanation for another interesting finding, originally reported by Coughlin and Patel in 1987. They presented structured and randomized forms of two different clinical cases (endocarditis and arteritis), and then asked the subjects to recall in writing a random version (of one case) and a structured version (of the other case). Students recalled correctly approximately the same proportions (@ 30%) of information from both normally ordered and randomized versions of the texts. Intriguingly, while in the case of arteritis experts recalled approximately the same amount of information regardless the structure of the case (45-50%), in the case of endocartitis the randomization of the data leaded to a significant decrease in recall performance (from @60% to 35-40%). Because the experimental condition was the same in both cases and we can not assume that different knowledge structures underlie the two illnesses, obviously the difference in memory performance reflects differences intervened in processing. According to Vicente and Wang (1998), the modification was caused by the alteration of the temporal order: contrary to arteritis, the symptoms of endocartitis have a specific intrinsic temporal order. The randomization disrupted this, and so, it eliminated one of the most important goal-relevant constraints. Experts’ skills, developed through gradual adaptation to the domain’s goal-relevant constraints, could not be used as efficiently as in situations similar to those in which they have acquired these skills. Although such a superior adaptation might seem very advantageous, highly specialized skill-set and domain knowledge can act as mental set, causing fixation, and so, producing inadequate or inefficient responses in non-routine, creative situations (Wiley, 1998).

            Strangely, Gilhooly & al. (1997) found that diagnostic accuracy in a ECG interpretation task did not correlate with accuracy of (incidental) recall though it would be reasonable to expect that the more accurate the diagnosis the more precise the recall will be. Considering on the one hand that, regarding diagnostic accuracy, an expertise effect was still observable, on the other hand that a correct diagnosis is realizable only with full (or, as complete as possible) knowledge of the illness in point, it seems that activation of knowledge about current diagnostic category will not solely improve memory performance. In order to find the explanation we have to direct our investigation to the analysis of different patterns in solution- and knowledge organization-strategies that experts’ use.

 

 

 

 

 

 

Patterns of reasoning: mixture or break-down?

 

Generally, in the medical domain, a series of study indicate that expert diagnostic reasoning is more accurate, efficient and context-sensitive that the reasoning of beginners. This is made possible by the fact that the physicians’ knowledge is gradually compiled into flexible, script-like structures that integrate both clinical and biomedical knowledge, as well as knowledge about “enabling conditions”. Expert reasoning - guided by such “illness-scripts” - rely on fast, pattern recognition-like processing, while novice reasoning is sealed by their rigid, isolated knowledge.

One of the major questions is whether expert physicians use forward reasoning like other domains’ experts (see the almost classic example from physics, Larkin, McDermott, Simon, and Simon, 1980) or use some form of hypothetico-deductive reasoning which is definitely closer to how novice try to reach the solution. According to Elstein, Shulman, and Sprafka (1978) experienced physicians act like less experienced subjects or novices (at least in the early phases of the problem solving process), generating a very limited number of tentative diagnoses. This makes experts and beginners to be more alike than different. However, more recent findings made necessary some revision of this apparent resemblance. Patel and Groen (1986) observed a rather odd relation (or correlation) between the direction of reasoning the accuracy of the diagnosis. They found that “the experts with accurate diagnoses used bottom-up forward reasoning whereas the experts with inaccurate diagnoses used at least some top-down backward reasoning” (pp.: 107). So, the author’s attempt to clear up which strategies are used under different conditions ended in formulating even more questions: is the inaccurate diagnosis due to the use of generate-and-test strategy? Could it be really possible that the directionality of the reasoning changes depending on specific factors? Or, are certain properties of the medical domain making necessary the use of backward reasoning, considering that, in contrast to other domains, in this case is quite difficult, if not impossible to elaborate a complete representation of the problem at the beginning of the investigation? If we want to understand the differences in the directionality of reasoning, we have to take in consideration both the benefits and the drawbacks of each reasoning pattern. In the case of the data-driven strategy hypotheses are generated from data and domain knowledge, which makes it highly error-prone, especially when the domain knowledge is inadequate and lacks consistency. The imperfect knowledge base manifests itself in an imperfect representation, which leads to the use of the backward strategy. In this case people use the input information in order to verify, or sometimes simply to justify their already existing hypotheses (Ericsson and Charness, 1994). Yet, backward chaining can prove unavoidable in some circumstances where data-driven processing is not applicable. For example, under uncertainty, where a part of the usually accessible and applicable rules can not be used, experts “turn back” to backward reasoning (Patel and Groen, 1986). The same way, forward reasoning is abandoned when unrelated facts are introduced in the description of the case. When experts encounter any unexplained findings, they have to test these informations against the main diagnosis, to account for these “loose ends”, and so they are forced to use backward strategy, or to be more precis: a mixture of the two strategies (Patel, Groen, and Arocha, 1990). We do not know for sure that data-driven processing and accurate diagnosis are tightly linked, but seems to be reasonable to accept that the breakdown of the forward reasoning can account at least partially for the inaccuracy of the diagnosis.

 

            A double-tracked road: applying biomedical and clinical knowledge

 

Given that in real-life setting a lot of necessary information is collected later during the diagnostic process (unless the problem is particularly easy), it is almost impossible to build a completely adequate representation, and without that pure forward reasoning is also impossible. For beginning few hypotheses are generated based on the presented signs and symptoms (forward reasoning), then these hypotheses are tested by searching for other indicators of each supposed disease (backward reasoning). This mixed pattern of reasoning implies combination of different type of information, namely biomedical and clinical knowledge. Biomedical knowledge concerns pathological mechanisms, processes underlying the manifestations of disease. This kind of information puts constraints upon the ways signs and symptoms are related, and it is used in generating elaborations. Szolovits, Patil, and Schwartz (1988) even suggested that biomedical knowledge limits somehow the number of hypothesis, and so protects clinical reasoning from becoming inefficient because of the overwhelming amount of information. Examining cardiologists’ think-aloud protocols Gilhooly & al. (1997) observed that this type of knowledge is mainly invoked in order to evaluate alternative hypotheses. Clinical knowledge refers to the manifestations of disease, like complaints, signs and symptoms, in short the attributes of the patient. Differences between experts and novices are revealed regarding not only the directionality of their reasoning but also the type of knowledge they are applying during diagnostic process (Gilhooly & al., 1997). Lesgold & al. (1988) emphasized the role of biomedical knowledge in medical reasoning, because experts applied better-suited, more detailed biomedical knowledge. Intriguingly, other authors reached completely different conclusions. Schmidt, Boshuizen, and Hobus (1988, cited after Boshuizen and Schmidt, 1992) reported that references to biomedical knowledge are almost missing in experts’ protocols. That is quite surprising because this is the kind of knowledge needed to reason forward. Besides, when students starts clinical internships they reason predominantly in terms of normal functions and abnormal processes, searching for explanations to how disturbances may result in diseases. A possible explanation accounting for the “disappearance” of biomedical knowledge from the clinical reasoning is offered by knowledge encapsulation. It is suggested that repeated knowledge application in similar situations might result in “abbreviations of search paths” (Boshuizen and Schmidt, 1992, p.: 176). That could also explain why experts’ performance in recall tasks is independent of time constraints on studying. So, in the process of mastering the structure of both type of knowledge will suffer considerable changes by gradual packaging. Lower level concepts and their relations are embedded, integrated in “high-level concepts with same explanatory power” (Schmidt and Boshuizen, 1993, p.: 347). Therefore reasoning do not “jumps” from the clinical description of the case to a diagnostic category, instead become increasingly efficient due to development of flexible cognitive means. In terms of the constraint attunement hypothesis (Vicente and Wang, 1998), the goal is to combine and interpret data in such manner that would lead to the selection of a single diagnostic category. Biomedical knowledge puts serious constraints on how pathological deviations from normal functioning manifest itself in disease, and so, associating signs and symptoms with underlying pathological mechanisms helps physicians to exclude many possibilities already in the initial phases or to distinguish between diseases (for example, in the case of diseases with very similar manifestations). Experts’ encapsulated knowledge could be interpreted as an efficient mean resulting from gradual attunement to the constraints imposed by knowledge about principles of normal functioning and abnormal deviations. This possibility gains significantly in plausibility if one considers that subjects at an intermediate level of expertise recall far more pathophysiological knowledge then experts or novices. Processing information by activating a detailed causal pathophysiological knowledge base seems to be an important stage in the development of expertise, ideal for repeated application of biomedical knowledge in many analogous situations. And that is the essential condition for finding those abbreviations of search paths mentioned earlier. [However, it is necessary to point out that some studies failed to reproduce the so-called intermediate effect (as it is the case of one reported by van den Wiel & al., 1998).]

 

Efficiency born from intertwined knowledge

 

It seems that the differences between expert and non-expert physicians’ performance are rather due to the differences in structuring experiences then to differences in general problem solving capacity (Proctor and Dutta, 1995). Moreover, as Grant and Marsden’s (1988) study suggested, expert performance relies (among other factors) on individual experiences rather then a common core of knowledge. We have to emphasize the qualitative aspects of the knowledge, because it seems that novices use the same amount, but not the same kind of knowledge as do expert physicians. Medical school provides beginners only with a limited and textbook-like common knowledge base, which need to be shaped by years of experience in order to reach expert-level performance.

We mentioned earlier that Elstein & al. findings suggest that there are no differences between medical experts and novices regarding the moment of generating first hypothesis, of which purpose’s is simply to constrain the search space and to reduce the demands on working memory. Nevertheless, there are serious differences on the quality of the diagnostic hypotheses generated, because the diagnostic accuracy increases if the initial set of working hypotheses contains the correct one (Barrows, Norman, Neufield, and Feightner, 1982, cited after Custers & al., 1996). In this case the correct solution is almost always recognized, which might suggest that the process of hypothesis generation based on the input data depends heavily upon how biomedical and clinical knowledge is organized in memory. Experience on how to handle contextual information (for example, age or hereditary influences) also plays an important role in the generation of initial diagnostic hypothesis (Hobus & al., 1987). The knowledge activated by the contextual information helps to eliminate from the beginning some improbable hypothesis and to admit more appropriate ones.

In contrast with other fields with more perceptually configured stimuli, like chess, in medical cognition it is improbable that simple pattern-recognition mechanisms or increased sensitivity to correlation between visible features can totally account for expert performance. It is suggested that standard recall paradigm is not a sufficient measure of medical expertise (Norman, Brooks, and Allen, 1989; Ericsson and Lehmann, 1996). Even in some special areas like radiology, dermatology or histology, the interpretation of the features as being normal or abnormal seems to guide both the search and the recognition of features, so that special knowledge structures are involved even in these visual categorization tasks.

In order to describe the complex knowledge structures involved in medical problem solving, Feltovich and Barrows (1984) introduced the notion of “illness script”.  Like schemas and mental models help us to deal with other kind of information, these structures are general knowledge frames that facilitate the construction of a representation of a particular patient’s medical problem. It should be seen as a narrative structure, consisting of three interrelated main components: 1.) enabling conditions; 2.) the faults; 3.) the consequences.

Enabling conditions, or contextual information regard factors that influence the probability of the acquisition of a specific disease (age, sex, risk behavior, previous medical history, etc.). In the early stages of the problem solving most of the information available are these, so it is understandable why they have such a great impact on the search for an answer. A certain medical history may act as constraint, reducing drastically the number of diagnostic alternatives, sometimes to a single one. In addition to their potential availability, enabling conditions have another important property: their probabilistic relationship to the presence of certain disease. The physician, knowing that a risk behavior augment the probability of a disease, will take into account and investigate possibilities which might not been indicated by other signs and symptoms. Hobus & al. (1987) found that experts produced almost 50% more correct hypotheses as compared to novices when they were given information about the patients’ medical history, because they were much better able to use these information in order to activate the proper script.

The faults, category that contains information about the major malfunctions, expressed in biomedical terms, concerns mainly the pathological mechanisms underlying the disease. Determining the alterations is essentially to distinguish between alternative diagnoses and to establish the proper treatment, considering that certain diseases have very similar surface manifestations (Gilhooly & al., 1997).

The category of consequences of the pathological factors includes the signs and the symptoms that build up the clinical image. A part of these information are also available in the early stages of the problem solving, but some of them have to be revealed by a series of questions or medical investigation, and this search is also guided by the experts’ initially activated knowledge and suppositions.

There is, however, some disagreement concerning the manner of how this illness scripts functions. Feltovich and Barrows claimed that such an illness script is constructed for each individual patient, yet researchers like Custers, Boshuizen and Schmidt (1996) extended Schank and Abelson’s original script idea, and sustained that these structures describe a general sequence of events, embedding biomedical, as well as clinical knowledge. Activated together as a whole by the information available in the initial phase, the illness script is instantiated by the data of the current case. This process may become a routine when the physician has to handle a slight amount of information or the patient presents all of the expected enabling conditions and consequences. The high flexibility of the diagnostic process might be explained by the fact that, due to activated illness-script, any new incoming information that is typical for the current script will be processed faster and with as little effort as possible. The atypical or inconsistent information can cause a disruption in the smoothness of the process, because it requires further elaboration, searching for explanations that would allow the preservation of the originally activated script. Unfortunately, confrontation with atypical cases and features increases the danger of making mistakes, because physician think mainly probabilistic, and the likelihood of an atypical case is subjectively underestimated, and that consequently could lead to the acceptance of a more typical, yet wrong diagnosis.

An interesting finding is that while experts make use of both enabling conditions and consequences as source of activation of diagnostic hypotheses, beginners seem to rely merely on the signs and symptoms presented by the patient. Moreover, contrary to the inexperienced ones, experienced physicians process more efficiently prototypical consequences if those are preceded by prototypical enabling conditions. This suggest that integration of the enabling conditions occur in the later stages of development of expertise providing expert physicians’ with a great sensitivity for typical cases, and so making possible for them to allocate more time and effort to the atypical ones (Custers & al., 1996).

The influence of different kind of information is also detectable in such visually based tasks like ECG or X-ray film interpretation. Both ECG traces and X-ray film can be characterized as uncontextualised information source because interpretation (and diagnoses) could be attained based solely on these materials. Gilhooly & al. (1997) described two routes by which the interpretation can be done (regardless of domain). The “surface-based” route relies mainly on the use of clinical knowledge; the signs detected in the traces are associated with a particular diagnostic category. The second route requires the use of biomedical knowledge; the diagnosis is established through identification of an underlying process that would generate the specific trace. Neither the former, nor the later route does suppose information regarding enabling conditions. Even so, diagnostic accuracy is still determined by expertise level, which suggest that ECG trace and X-ray films provide sufficient clinical information for correct diagnosis. Although X-ray films torn out from their context may be objects of interpretation, there are at least two aspects of clinical information not present in the stimuli, which might improve radiographic inspection: 1.) indication of specific location for intensive evaluation (for example, the patient complaint about pain in a certain area), and 2.) clues to search for particular abnormalities (Berbaum & al., 1988). Norman, Brooks, Coblentz, and Babcook (1992) presented similar, though more detailed results. They supplemented X-rays films from confirmed normal or bronchiolitis patients with normal or abnormal clinical histories. Some of the films were definitely normal or abnormal, while others were formerly judged as being equivocal or ambiguous. The subjects’ task was to state the presence or the absence of five features, and then to rate the likelihood of bronchiolitis. The results confirmed that prior biasing information, regardless of the expertise level, could influence diagnostic judgments and feature ratings, which means that features not act as an independent source of information. Experts were biased toward identifying a certain feature on normal films accompanied by positive history for bronchiolitis. (In the case of normal films presented with normal histories, the number of films on which they detected the same feature was smaller then the former.) The interpretation of X-ray films is constrained by consistency of signs and symptoms gathered both from film and history, as well as knowledge about possible underlying pathological mechanisms. But, by associating normal films with abnormal history some of the goal-oriented constraints are violated, because signs described in history not correlate with features on the film. This creates uncertainty especially if the films are radiologically equivocal, because signs provided by histories were not consistent with the signs present on films. This made difficult the discrimination between normal variation and abnormal features, and leaded to preference for abnormal identifications. (As one of my friends said: “it is much easier to find a disease then to accept that the person is healthy”).

 

Final remarks

 

Generally, experimental data show that medical expert are capable of outstanding accuracy in diagnosis; forward reasoning seems to be the key of their success, but this data-driven process breaks down in several situations (because of the missing information, or encountering unaccounted data, as well as under uncertainty). In order to explain any inconsistency in data, they have to use backward strategies, which seems to be associated with the decrease of diagnostic accuracy, the main criterion for medical expertise. But, they also develop powerful cognitive means, like encapsulated knowledge or illness scripts; the huge amount of knowledge necessary for mastering in medicine is shaped and reshaped with every stage in the process of development of true expertise. Still, there are many gaps in our knowledge about medical problem solving, many aspects of expert performance, insufficiently studied so far, seems to remain a question of intuition and subconscious processing.

Understanding the principles and rules which governs expert problem solving will make possible new advances in artificial intelligence. Collaboration between cognitive scientist will provide improvement and development of both new expert systems and decisions making support-systems, and will allow optimization of medical monitoring technology. But, far more important applications concerns instruction. Based on a theory of medical expertise, flexible methods for facilitating reasoning process for medical students can be worked out. The beneficiaries will be not only the students, but also the experts that we are learning from today. Considering that “prior knowledge critically shapes one’s perception of new information” (Patel, 1998, p.: 94), both AI and medicine school will have to develop means that would allow experts to continuously update and integrate new information.

 

References

 

Berbaum, K.S., El-Khoury, G.Y., Franken, E.A., Kathol, M., Montgomery, W.J., & Hesson, W. (1988). Impact of clinical history on fracture detection with radiography. Radiology, 168, 507-511.

Boshuizen, H.P.A., & Schmidt, H.G. (1992). On the role of biomedical knowledge in clinical reasoning by experts, intermediates and novices. Cognitive Science, 16, 153-184.

Brooks, L.R., Norman, G.R., & Allen, S.W. (1991) Role of specific similarity in a medical diagnostic task. Journal of Experimental Psychology: General, 120(3), 278-287.

Chase, W.G., & Ericsson, K.A. (1981). Skilled memory. In J.R. Anderson (Ed.), Cognitive skills and their acquisition, pp. 141-189. Hillsdale, NJ: Erlbaum.

Chase, W.G., & Simon, H.A. (1973). Perception in chess. Cognitive Psychology, 4, 55-81.

Claessen, H. F., & Boshuizen, H. P. A. (1985). Recall of medical information by students and doctors. Medical Education, 19, 61-67.

Custers, E.J.F.M., Boshuizen, H.P.A., & Schmidt, H.G. (1996). The influences of medical expertise, case typicality, and illness script component on case processing and disease probability estimates. Memory & Cognition, 24(3), 384-399.

Elstein, A. S., Shulman, L. S., & Sprafka, S. A. (1978) Medical problem solving. An analysis of clinical reasoning. Harvard University Press, Cambridge, MA.

Ericsson, K.A. (1998). The scientific study of expert levels of performance: general implications for optimal learning and creativity [1]. High Ability Studies, 9(1), 75-100.

Ericsson, K.A. & Charness, N. (1994). Expert performance: Its structure and acquisition. American Psychologist, 49, 725-747.

Ericsson, K.A., & Lehmann, A.C. (1996). Expert and exceptional performance: Evidence of maximal adaptation to task constraints. Annual Review of Psychology, 47, 273-305.

Feltovich, P. J., & Barrows, H. A. (1984). Issues of generality in medical problem solving. In H. G. deVolder and H. G. Schmidt (Eds.), Tutorials in problem based learning. Assen, Holland: van Gorcum.

Gilhooly, K.J., McGeorge, P., Hunter, J., Rawles, J.M., Kirby, I.K., Green, C., & Wynn, W. (1997). Biomedical knowledge in diagnostic thinking: The case of electrocardiogram (ECG) interpretation. European Journal of Cognitive Psychology, 9(2), 199-223.

Grant, J., & Mardsen, P. (1988). Primary knowledge, medical education and consultant expertise. Medical Education, 22, 173-179.

Gobet, F. (1994). Memory in chess players: chunks, schemata, or both? (Complex Information Processing Working Paper No. 517). Pittsburgh, PA: Department of Psychology, Carnegie Mellon University.

Hobus, P.P.M., Schmidt, H.G., Boshuizen, H.P.A., & Patels, V.L. (1987). Contextual factors in the activation of first diagnostic hypotheses: expert-novices differences. Medical Education, 21, 471-476.

Kerbeshian, L. (1991). Mediator use during inferential problem solving task: Medical diagnosis. In M.L. Languis, P.J. Naour, J.J. Buffer (Eds.), Cognitive science. Contribution to educational practice, pp. 380-396. Gordon and Breach Science Publushers, Philadelphia.

Larkin, J. H., McDermott, J., Simon, D. P., & Simon, H. A. (1980). Expert and novice performance in solving physics problems. Science, 208, 1335-1342.

Lesgold, A., Rubinson, H., Feltovich, P.J., Glaser, R., Klopfer, D., & Wang, Y. (1988). Expertise in a complex skill: Diagnosing X-ray pictures. In M.T.H. Chi, R. Glaser, and M. Farr (Eds.), The nature of expertise. Hillsdale, NJ: Erlbaum.

Mérő, L. (1997). Észjárások. A racionális gondolkodás korlátai és a mesterséges intelligencia. Tericum Kiadó, Budapest.

Myles-Worsley, M., Johnston, W.A., & Simons, M.A. (1988). The influence of expertise on X-ray image processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(3), 553-557.

Norman, G.R., Brooks, L.R., & Allen, S.W. (1989). Recall by expert medical practitioners and novices as a record of processing attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15(6), 1166-1174.

Norman, G.R., Brooks, L.R., Coblentz, C.L., & Babcook, C.J. (1992). The correlation of feature identification and category judgments in diagnostic radiology. Memory & Cognition, 20(4), 344-355.

Patel, V.L. (1998). Individual to collaborative cognition: a paradigm shift? Artificial Intelligence in Medicine, 12, 93-96.

Patel, V.L., & Groen, G.J. (1986). Knowledge based strategies in medical reasoning. Cognitive Science, 10, 91-116.

Patel, V. L., Groen, G. J., & Arocha, J F. (1990). Medical expertise as a function of task difficulty. Memory & Cognition, 18(4), 394-406.

Pople, H.E. (1982). Heuristic methods for imposing structure in ill structured problems: The structuring of medical diagnostics. In P. Szolovits (Ed.), Artificial intelligence in medicine, pp. 119-190. Boulder, CO: Westview Press.

Proctor W.R. & Dutta, A. (1995). Skill acquisition and human performance. ADT, Sage Publications, London.

Schmidt, H.G., & Boshuizen, H.P.A. (1993). On the origin of intermediate case recall. Memory & Cognition, 21(3), 338-351.

Szolovits, P., Patil, R.S., & Schwartz, W.B. (1988). Artificial intelligence in medical diagnosis. Annals of internal medicine, 108(1), 80-87.

Vicente, K.J., & Wang, J.H. (1998). An ecological theory of expertise effects in memory recall. Psychological Review, 105(1), 33-57.

van del Wiel,  M.W.J., Schmidt, H.G., & Boshuizen, H.P.A. (1998). A failure to reproduce the intermediate effect in clinical case recall. Academic Medicine, 73, 894-900.

Wiley, J. (1998). Expertise as mental set: The effects of domain knowledge in creative problem solving. Memory & Cognition, 26(4), 716-730.