Medical errors as a result of specializationIntroductionThe cost of expertiseAccuracy of recallReduced flexibilityCharacteristics of medical expertiseDiagnostic reasoning modeExperts working outside their domainThe present workMethodsParticipantsMaterialDesign and procedureResultsVerbal protocol analysesGeneration of diagnostic hypothesesCues used in hypothesis generationAnomaliesSummary of protocol analysis resultsANOVA results for the probabilities assigned to different specialtiesGeneralized linear modelModeling the probability assigned to the correct specialty (P(CS))Modeling the probability assigned to the correct diagnosis (P(CD))Modeling the probability assigned to the participant’s own specialty (P(OS))Summary of the generalized linear model resultsDiscussionReferencesMedical errors as a result of specializationAhmad Hashem,a,*Michelene T.H. Chi,band Charles P. FriedmancaHealthcare Industry Solutions Group, Microsoft Corporation, One Microsoft Way, Redmond, WA 98052, USAbLearning Research and Development Center, University of Pittsburgh, 3939 O’Hara Street, Pittsburgh, PA 15260, USAcCenter for Biomedical Informatics, University of Pittsburgh, Forbes Tower, Suite 8084 200 Lothrop Street, Pittsburgh, PA 15213-2582, USAReceived 8 July 2003AbstractErrors in medicine result in over 44,000 preventable deaths annually. Some of these errors are made by specialized physicians atthe time of diagnosis. Building on error frameworks proposed in the literature, we tested the experimental hypothesis that physicianswithin a given specialty have a bias in diagnosing cases outside their own domain as being within that domain. Thirty-two board-certified physicians from four internal medicine subspecialties worked four patient cases each. Verbal protocol analysis and generallinear modeling of the numerical data seem to confirm the experimental hypothesis, indicating that specialists try to ‘‘pull’’ casestoward their specialty. Specialists generate more diagnostic hypotheses within their domain than outside, and assign higherprobabilities to diagnoses within that domain. 2003 Published by Elsevier Inc.Keywords: Medical errors; Diagnostic errors; Cost of expertise1. IntroductionBetsy Lehman, a Boston Globe health reporter, diedfrom an overdose during chemotherapy. Willie Kinghad the wrong leg amputated. Ben Kolb was eight yearsold when he died during surgery due to a drug mixup[1,2]. And these are just the ‘‘tip of the iceberg’’ ofmedical errors [1,2].One important type of medic al error occurs at thetime of diagnosis. The popular press cites cases such asthat of Dr. Franklin K. Yee, whose abdominal pain wasdiagnosed as viral gastroenteritis by a gastroenterolo-gist, caused him to be admitted to a coronary care unitby a cardiologist, was suspected by a nephrologist to bethe result of kidney stones, and eventually was found onabdominal surgery to be the result of a ruptured ap-pendix [3]. This phenomenon of different specialistsprojecting their specialties on a patient has not beenstudied systematically. In the present paper, the role ofmedical specialization in inducing biases that may un-derlie some diagnostic medical errors is investigatedempirically.2. The cost of expertiseIt may seem strange to talk about the costs of beingan expert, but there is increasing experimental evidencethat the benefits of expertise are not without costs. Thecosts of expert ise can be divided into two main catego-ries: Those related to accuracy of recall, and those re-lated to inflexibility.2.1. Accuracy of recallExperts may outperform novices in recalling the de-tails of a problem or text. But when the domainknowledge of experts cannot be utilized, experts tend tounderperform novices. For example, in a study on recallof random chessboard positions, the performance ofchess experts was slightly worse than that of novices [4].Similarly, in a study on memory for baseball texts,participants with high baseball knowledge recalled sig-nificantly less baseball-irrelevant propositions from a*Corresponding author. Fax: 1-425-936-7329.E-mail address: [email protected] (A. Hashem).www.elsevier.com/locate/yjbinJournal of Biomedical Informatics 36 (2003) 61–691532-0464/$ - see front matter  2003 Published by Elsevier Inc.doi:10.1016/S1532-0464(03)00057-1text passage describing part of a baseball game than didparticipants with low baseball knowledge [5].2.2. Reduced flexibilitySome of the early work on the costs of expertise wasdone by Ggestalt psychologists [6–8]. Luchins arguedthat experience produced an Einstellung, or mental set,that limited the search space of subsequent problemsolving. More recent studies have shown similar results.For example, Shiffri n and Schneider [9] reported thatpeople trained for several thousand trials to detect visualtargets among distractors in a consistent- mapping con-dition were at a serious disadva ntage when the targetand distractor sets were reversed. Wiley [10] showed thatproblem solvers with a large amount of domainknowledge are confined by their knowledge to one areaof the search space—a condition that is efficient if thesolution happens to fall in this area but which backfiresif it does not.3. Characteristics of medical expertiseThere is no reason to suspect that exp ertise in medi-cine is different from expertise in any other domain. Infact, research results have emphasized similarities (e.g.[11–15]). The general characteristics of expertise aresummarized elsewhere [16,17] and will not be repeatedhere. Some characteristics of expertise in the medicaldomain that bear relevance to the present study will bediscussed.3.1. Diagnostic reasoning modeAccording to Patel and Groen [14], expert cliniciansconfronted with routine cases use a data-driven (for-ward) approach, in which diagnoses are generated fromdata by applying a small set of if/then production ruleswithout generating intermediate hypotheses and evalu-ating them. Less expert clinicians, such as medical stu-dents or residents, on the other hand, tend to use ahypothesis-driven (backward) ap proach, in which rea-soning occurs backwards from a hypothesis in an at-tempt to find data that elucidates it [18]. This assertion isconsistent with findings in other domains such asphysics [19] and mathematics [20], and with the generalnotion that in routine situations experts tend to usehighly specific problem-solving structures [21].3.2. Experts working outside their domainCognitive literature