Event-related potentials reveal the relations between feature representations at different levels of abstraction.

In this paper, we use behavioural methods and event-related potentials (ERPs) to explore the relations between informational and instantiated features, as well as the relation between feature abstraction and rule type. Participants are trained to categorize two species of fictitious animals and then identify perceptually novel exemplars. Critically, two groups are given a perfectly predictive counting rule that, according to Hannah and Brooks (2009. Featuring familiarity: How a familiar feature instantiation influences categorization. Canadian Journal of Experimental Psychology/Revue Canadienne de Psychologie Expérimentale, 63, 263-275. Retrieved from http://doi.org/10.1037/a0017919), should orient them to using abstract informational features when categorizing the novel transfer items. A third group is taught a feature list rule, which should orient them to using detailed instantiated features. One counting-rule group were taught their rule before any exposure to the actual stimuli, and the other immediately after training, having learned the instantiations first. The feature-list group were also taught their rule after training. The ERP results suggest that at test, the two counting-rule groups processed items differently, despite their identical rule. This not only supports the distinction that informational and instantiated features are qualitatively different feature representations, but also implies that rules can readily operate over concrete inputs, in contradiction to traditional approaches that assume that rules necessarily act on abstract inputs.

Influence of familiar features on diagnosis: instantiated features in an applied setting.

Medical diagnosis can be viewed as a categorization task. There are two mechanisms whereby humans make categorical judgments: "analytical reasoning," based on explicit consideration of features and "nonanalytical reasoning," an unconscious holistic process of matching against prior exemplars. However, there is evidence that prior experience can also operate at the level of individual "instantiated" features (Brooks & Hannah, 2006). The present studies examined instantiated features in medical diagnosis. Four "pseudopsychiatric" conditions, each described by four characteristic features, were taught to undergraduate psychology students. They practiced on additional cases, then were tested on new cases with features from two conditions. In Experiment 1, diagnoses associated with familiar features presented one or three times during practice were assigned a higher probability than those with novel features. Experiment 2 showed that the impact of feature frequency was dependent on its consistency with the case diagnosis. Experiment 3 showed that the effect of feature familiarity was not confined to cases with two equiprobable diagnoses. Experiment 4 showed that the effect remained after a 24 hour delay. These four studies demonstrated that features seen in practice have a greater influence on diagnosis than novel synonyms. In fact, seeing a feature once within the appropriate context (a patient case in which it is a member of the primary diagnosis) was sufficient to form a diagnostic association equivalent to instantiations seen four times in a different context. The results of these studies have implications for theories of categorization and for teaching clinical reasoning.

Teaching posttraining: influencing diagnostic strategy with instructions at test.

It is believed that medical diagnosis involves two complementary processes, analytic and similarity-based. There is considerable debate as to which of these processes defines diagnostic expertise and how best to teach clinical diagnosis and reduce diagnostic errors. The purpose of these studies is to document the use of these strategies in medical students. We shifted the balance in use of these processes and improved diagnostic accuracy with instructions given posttraining at the moment of diagnosis. Analytic processing reflecting the degree to which cases contain the diagnostic rules was indexed by the rate of accuracy on typical versus atypical cases (typicality effect). Similarity-based processing reflecting the degree to which cases resemble previously encountered cases was indexed by the rate of accuracy on similar versus dissimilar cases (similarity effect). Two studies are presented illustrating that diagnosis involves the coordination of analytic and similarity-based processes and that differential instruction given at test shifts the balance in the use of these processes. Study 1 illustrated that participants adopting an analytic strategy exhibit a larger effect of typicality. Participants adopting a similarity-based strategy exhibit a larger effect of similarity. The diagnostic approach of students given no instructions was predominantly analytic. Dual instructions in which participants first employed similarity-based processing followed by the application of rules improved overall accuracy. Study 2 investigated two versions of dual instructions and illustrated that assessing a case with the rules of diagnosis first may inhibit the subsequent use of similarity-based reasoning. The implications for diagnostic expertise and pedagogy are discussed.

The influence of familiar non-diagnostic information on the diagnostic decisions of novices.

CONTEXT: Previous research has demonstrated the influence of familiar symptom descriptions and entire case similarity on diagnostic reasoning. In this paper, we extend the role of familiarity to examine the influence of familiar non-diagnostic patient information (e.g. name and age) on the diagnostic decisions of novices, both immediately following training and after a delay. If an instance model (reliance on similar previously seen cases) has strong explanatory power in clinical reasoning, we should see an influence of familiar patient information on later cases containing similar identifying characteristics even though such information is objectively irrelevant. METHODS: Thirty-six participants (undergraduate psychology students) were trained to competence on four simplified psychiatric diagnoses and allowed to practise their diagnostic skills on 12 prototypical case vignettes, for which feedback was provided. One-third of participants were tested immediately, one-third following a 24-hour delay, and one-third following a 1-week delay; all were tested on novel cases. Test cases were created to have two equiprobable diagnoses, both of which were supported by two novel symptom descriptions. However, one diagnosis was also supported by non-diagnostic patient information similar to information on a patient seen in the training phase. A deviation from an equal assignment of diagnostic probability, in support of the familiar patient information, demonstrates a reliance on the familiar, non-diagnostic information, and therefore indicates an instance model of reasoning. RESULTS: Participants assigned significantly higher diagnostic probability to the diagnosis cued by the familiar patient information (52.6%) than to the plausible alternative diagnosis (38.9%). Participants also reported a higher number of clinically relevant symptoms to support the diagnosis associated with the familiar patient information than to support the plausible alternative diagnosis. The influence of familiar patient identity was consistent across delay periods and cannot be accounted for by the forgetting of diagnostic rules. CONCLUSIONS: Participants were clearly relying on familiar patient identity information as evidenced by their diagnostic conclusions and differential reporting of clinically relevant features. These results support an instance model of reasoning which is not limited by whole case similarity or similarity of diagnostic information.

Featuring familiarity: how a familiar feature instantiation influences categorization.

We demonstrate that a familiar looking feature can influence categorization through 2 different routes, depending on whether a person is reliant on abstract feature representations or on concrete feature representations. In 2 experiments, trained participants categorized new category members in a 3-step procedure: Participants made an initial categorization, described the rule-consistent features indicated by the experimenter, and then recategorized the item. Critical was what happened on the second categorization after participants initially categorized an item based on a familiar, but misleading, feature. Participants who were reliant on abstract features most commonly reversed themselves after the rule-consistent features were pointed out, suggesting that the familiar feature had biased attention. Participants who were reliant on concrete feature representations, however, most commonly persisted with the initial response as if the familiar feature were more important than its rivals-the familiar feature biased decision making.

Teaching from the clinical reasoning literature: combined reasoning strategies help novice diagnosticians overcome misleading information.

OBJECTIVE: Previous research has revealed a pedagogical benefit of instructing novice diagnosticians to utilise a combined approach to clinical reasoning (familiarity-driven pattern recognition combined with a careful consideration of the presenting features) when diagnosing electrocardiograms (ECGs). This paper reports 2 studies demonstrating that the combined instructions are especially valuable in helping students overcome biasing influences. METHODS: Undergraduate psychology students were trained to diagnose 10 cardiac conditions via ECG presentation. Half of all participants were instructed to reason in a combined manner and half were given no explicit instruction regarding the diagnostic task. In Study 1 (n = 60), half of each group was biased towards an incorrect diagnosis through presentation of counter-indicative features. In Study 2 (n = 48), a third of the test ECGs were presented with a correct diagnostic suggestion, a third with an incorrect suggestion, and a third without a suggestion. RESULTS: Overall, the instruction to utilise a combined reasoning approach resulted in greater diagnostic accuracy relative to leaving students to their own intuitions regarding how best to approach new cases. The effect was particularly pronounced when cases were made challenging by biasing participants towards an incorrect diagnosis, either through mention of a specific feature or by making an inaccurate diagnostic suggestion. DISCUSSION: These studies advance a growing body of evidence suggesting that various diagnostic strategies identified in the literature on clinical reasoning are not mutually exclusive and that trainees can benefit from explicit guidance regarding the value of both analytic and non-analytic reasoning tendencies.

Diversion of attention in everyday concept learning: identification in the service of use.

Many people tend to believe that natural categories have perfectly predictive defining features. They do not easily accept the family resemblance view that the features characteristic of a category are not individually sufficient to predict the category. However, common category-learning tasks do not produce this simpler-than-it-is belief. If there is no simple classification principle in a task, the participants know that fact and can report it. We argue that most category-learning tasks in which family resemblance categories are used fail to produce the everyday simpler-than-it-is belief because they encourage analysis of identification criteria during training. To simulate the learning occurring in many natural circumstances, we developed a procedure in which participants' analytic activity is diverted from the way in which the stimuli are identified to the use to which the stimuli will be put. Finally, we discuss the prevalence of this diverted analysis in everyday categorization.

It all make sense: biomedical knowledge, causal connections and memory in the novice diagnostician.

Although there is consensus among medical educators that students must receive training in the biomedical sciences, little is known regarding the role of biomedical knowledge in diagnosis. The present paper presents two studies examining the role of biomedical knowledge, specifically knowledge of causal mechanisms, in novice diagnosticians. In Experiment 1, two groups of participants are taught to diagnose a series of artificial diseases. In the causal learning condition students learn the underlying causal mechanisms for each feature. A second group learns the same diseases without the causal explanations. Participants are asked to diagnose a series of written cases immediately after training and again 1 week later. The results show that students who learn a causal model are better able to retain their diagnostic performance over time (89% correct vs. 78%). This finding is investigated further in Experiment 2, demonstrating that students rely more on casual information after a delay (mean probability of 57% vs. 43%). Together, the studies suggest that knowledge of underlying causal mechanisms can aid student memory for diagnostic categories and that use of causal knowledge changes over time.

The benefits of flexibility: the pedagogical value of instructions to adopt multifaceted diagnostic reasoning strategies.

OBJECTIVES: Building on the advice of previous research to avoid parsing diagnostic strategies too finely, recent studies have shown that teaching novices to utilise analytic and non-analytic reasoning strategies yields higher diagnostic accuracy than teaching either in isolation. This study assesses the extent to which students spontaneously adopt a combined approach and compares its benefits with those experienced with a contrastive learning strategy known to enhance analogical transfer. METHODS: A sample of 48 naïve students were trained to identify features on electrocardiograms (ECGs) and assign diagnoses. Half the participants learned in a standard manner, encountering diagnoses (and their associated features) in sequence. The remaining participants were explicitly instructed to draw comparisons between the diagnostic category being learned and another confusable diagnostic category (contrastive learning). Half the participants in both groups were further instructed to carefully identify all features while trusting guidance provided by feelings of familiarity (a combined reasoning strategy). The remaining participants were given no instructions on how to approach the diagnostic task. RESULTS: Greater diagnostic accuracy was achieved following both contrastive learning and instructions to use a combined reasoning strategy relative to the control conditions. These variables did not interact with each other, nor did they interact with novelty of the test case. The effects were observed immediately after learning and following a 1-week delay. DISCUSSION: The results emphasise the importance of explicitly empowering students to utilise multiple diagnostic strategies, including non-analytic approaches. In addition, this study reveals the benefit that can be gained from contrastive learning in a medical domain.

The role of biomedical knowledge in diagnosis of difficult clinical cases.

Although biomedical knowledge is believed to be of little value in diagnosis of routine clinical cases, studies of clinical reasoning have found that physicians revert to use of basic biomedical knowledge when faced with challenging clinical problems. The current paper presents two experiments that empirically examine the role of biomedical knowledge in diagnosis of difficult cases by novice diagnosticians. Novices are taught to diagnose a series of artificial diseases using either knowledge of causal mechanisms or a list of clinical features. In Experiment 1, participants are then tested on two types of clinical challenges: (1) case summaries with irrelevant findings; (2) cases using unfamiliar terminology. Participants with an understanding of underlying mechanisms out performed their counterparts on both types of cases. In Experiment 2, participants are tested 1 week after initial training. Participants with knowledge of causal mechanisms were found to do better on cases with unfamiliar terminology. The results of the two studies provide additional support for the critical role of biomedical knowledge in diagnosis of difficult clinical cases.

Producing biased diagnoses with unambiguous stimuli: The importance of feature instantiations.

In this article, the authors demonstrate a laboratory analogue of medical diagnostic biasing (V. R. LeBlanc, G. R. Norman, & L. R. Brooks, 2001) in 2 experiments and explore the basis of this effect. Before categorizing novel exemplars, participants first evaluated the likelihood that the item was a member of the category suggested on that trial: either the correct category or a plausible alternative category. This was sufficient to produce a substantial bias toward the suggested category despite the use of unambiguous stimuli, explicit rules, and unhurried conditions--each of which would be likely to limit diagnostic bias. The authors argue that the production of this effect requires distinguishing between particular feature instantiations and more abstract representations of those features as well as allowing people to adopt a particular decision strategy mediating the use of instantiated features: a feature-recognition heuristic.

Speed kills? Speed, accuracy, encapsulations and causal understanding.

BACKGROUND: The role of basic science, which provides causal explanations for clinical phenomena in medical education, is poorly understood. Schmidt has postulated that expert clinicians maintain this knowledge in 'encapsulated' form, indexed by words or phrases describing the processes. In the present paper we show that students who learn causal explanations have a more coherent understanding of the relation between diseases and clinical features which, in turn, influences recognition of words or phrases describing 'encapsulated knowledge' and the ability to maintain performance under speeded conditions. HYPOTHESES: In comparison to students who simply learn the features of 4 diagnostic categories, students who learn a causal explanation will: (a) recognise words describing encapsulated knowledge more accurately and (b) maintain or improve diagnostic performance under speeded conditions. METHODS: Two studies were conducted involving 4 'pseudo-endocrinology' diseases and undergraduate psychology students. One group learned signs and symptoms alone; the second group also learned a causal explanation. In study 1, they were then given a recognition memory task. In study 2, they were asked to diagnose new cases either (i) as quickly as possible or (ii) taking their time. RESULTS: In study 1, while there was no difference in recognising old words (90% versus 91%), the causal group was better able to recognise encapsulated and novel consistent words (50% versus 41%) (P = 0.02). In study 2 there was an interaction; causal students performed better under speeded conditions (71% versus 66%) but worse under thoroughness conditions (67% versus 73%), as predicted. CONCLUSIONS: Causal understanding leads to more coherent understanding of clinical conditions, which in turn leads to expert-like behaviour.

Instantiated features and the use of "rules.".

Classification "rules" in expert and everyday discourse are usually deficient by formal standards, lacking explicit decision procedures and precise terms. The authors argue that a central function of such weak rules is to focus on perceptual learning rather than to provide definitions. In 5 experiments, transfer following learning of family resemblance categories was influenced more by familiar-appearing features than by novel-appearing features equally acceptable under the rule. This occurred both when rules were induced and when rules were given at the beginning of instruction. To model this and other phenomena in categorization, features must be represented on 2 levels: informational and instantiated. These 2 feature levels are crucial to provide broad generalization while reflecting the known peculiarities of a complex world.

Giving learners the best of both worlds: do clinical teachers need to guard against teaching pattern recognition to novices?

PURPOSE: There has been much debate in the medical education literature regarding the extent to which feature-driven and nonanalytic (similarity-based) reasoning strategies define expertise, but the relative value of teaching these strategies, together or in isolation, remains uncertain. The purpose of this study was to compare the diagnostic accuracy achieved upon receiving instruction to use each strategy in isolation to that of a combined approach. METHOD: In 2003-04, 48 undergraduate psychology students from McMaster University in Ontario, Canada, were taught to diagnose ten cardiac disorders (including normal) via electrocardiogram (ECG) presentation. Twelve students were instructed to carefully identify all features present before assigning a diagnosis (feature first). Twelve were given the same instruction with notice that some test ECGs had been seen during training (implicit combined). Twelve were simply instructed to trust familiarity and diagnose based on this impression (similarity-based). Finally, 12 students were given feature first and similarity-based instructions in combination (explicit combined). RESULTS: No difference in diagnostic accuracy was observed between the groups given the feature first (42%) and first impression (41%) instructions (p > .4), but the groups instructed to use both strategies (explicitly or implicitly) performed significantly better (56% and 53%, respectively; p < .01). CONCLUSIONS: The results support an additive model of clinical reasoning in which instructions to be feature oriented and to trust similarity improve performance in novice diagnosticians.

The blossoms and the weeds.

As attested by the articles in this issue and the papers given at the festschrift, I have had the privilege of working with a talented group of people. In this article, a critique is given on the bouquet of themes that have bound many of us together, sorting the blossoms from some of the weeds. A brief description is then given of how these themes are working out in our current research on categorization.

The value of basic science in clinical diagnosis: creating coherence among signs and symptoms.

BACKGROUND: We investigated whether learning basic science mechanisms may have mnemonic value in helping students remember signs and symptoms, in comparison with learning the relation between symptoms and diagnoses directly. PURPOSE: To compare 2 approaches to learning diagnosis: learning how features of various conditions relate to underlying pathophysiological mechanisms and learning the conditional probabilities of features and diseases. METHODS: Undergraduate students (n = 36) were taught 4 disorders (upper motor neuron lesion, lower motor neuron lesion, neuromuscular junction disease and muscular disease), either using basic science explanations or (symptom x disease) probabilities. They were tested with diagnostic cases immediately after learning and 1 week later. RESULTS: On the immediate test, there was no difference in the results. One week later, the accuracy of the mechanism group remained at 0.52, but the performance of the probability group had dropped to 0.43. CONCLUSIONS: Knowledge of basic science may have value in clinical diagnosis by helping students recall or reconstruct the relationships between features and diagnoses.

Using comprehensive feature lists to bias medical diagnosis.

Clinicians routinely report fewer features in a case than they subsequently agree are present. The authors report studies that assess the effect of considering a more comprehensive description than physicians usually offer. These comprehensive descriptions were generated from photographs of dermatology and internal medicine and were complete and accurate. Groups of clinicians of varying expertise were asked to offer a diagnosis based solely on the comprehensive verbal description. This initial exercise decreased the subsequent diagnostic acumen of experienced participants with the photographs relative to a group that initially diagnosed from the photographs. Reasons that the initial consideration of a list of features, all of which are present in the photograph, would decrease diagnostic accuracy are discussed.

Limiting the playing field: does restricting the number of possible diagnoses reduce errors due to diagnosis-specific feature identification?

INTRODUCTION: Diagnostic hypotheses influence the identification of clinical features by medical trainees. This influence is strong enough to lead students to interpret features incorrectly if the initial diagnostic suggestion is incorrect. In the present study, we investigated whether reducing the pool of possible diagnoses at the time of test to a few highly plausible alternatives would focus the search for and interpretation of clinical features on a few alternative diagnoses and, as a result, reduce the influence of an initial diagnostic hypothesis on feature identification. METHODS: Naive students were taught 10 electrocardiographic (ECG) diagnoses. At test, they were asked to report all features visible on new ECGs. The test ECGs were presented with the suggestion of a tentative diagnosis (either the correct diagnosis or a plausible alternative) under 2 conditions: students were either instructed that the ECG represented one of 3 possible diagnoses (which were explicitly mentioned), or they were instructed that the ECG might represent any of the 10 diagnoses learned. RESULTS: Students' identification of the ECG features was strongly influenced by the diagnostic suggestion. Reducing the number of alternatives available at the time of test did not reduce the impact of a diagnostic suggestion on feature interpretation. DISCUSSION: Increasing the salience of alternative hypotheses does not reduce the impact of a diagnostic suggestion on the interpretation of clinical features.

The privileged status of prestigious terminology: impact of "medicalese" on clinical judgments.

PURPOSE: Health professionals frequently use medical terminology like dyspnea or nasopharyngitis. These two studies examine how the use of medical terms affects the judgments of seriousness, prevalence, and disease; and diagnostic judgments. METHOD: In study 1, a survey containing the names of 22 diseases with either a medical or lay description was completed by 47 undergraduate psychology students and 25 medical students, who were asked to judge seriousness, prevalence, and how "disease-like" it was. In study 2, undergraduate students learned four "pseudopsychiatry" conditions, each with four associated features. Features were presented in lay or medical versions. They were then tested with 18 new cases with two medical features from one condition and two lay terms from the other. RESULTS: In study 1, the medical students rated conditions as more disease-like, more serious, and less prevalent than did the psychology students. Medical descriptions were seen as significantly less common and somewhat more serious and more disease-like. In study 2, the participants rated the condition with medical features consistently more likely than the alternative, regardless of training condition. CONCLUSIONS: The specific words used to describe a feature or condition can have an impact on judgments of likelihood of disease, and, to a lesser extent, judgments of seriousness.

Practice makes perfect: the critical role of mixed practice in the acquisition of ECG interpretation skills.

PURPOSE: To examine the effect of instructional format on medical students' learning of ECG diagnosis. METHOD: Two experiments employed different learning and practice methods. In the first, students were randomly allocated to one of two instructional approaches, one organized around features (e.g., QRS voltage) and the other around diagnostic categories (e.g., bundle branch blocks), followed by a practice phase. In the second experiment, the instruction was standardized, and students were randomly allocated to one of two practice phases, either "contrastive" where examples from various categories are mixed together, or "non-contrastive" where all the examples in a single category are practiced in a single block. RESULTS: In the first experiment, there was no significant differences in students' diagnostic accuracy on novel ECG examples. In the second experiment, students exposed to the contrastive approach in the practice phase had superior diagnostic accuracy (46%) compared to 30% accuracy for the non-contrastive session, p < 0.05). CONCLUSION: These experiments highlight two important features in the design of instructional materials. First, learning around the features of the problem (analogous to problem-based learning) may have no advantages over learning the category. Second, the design and organization of deliberate practice can result in significant learning gain.