A Preventable Mode of Diagnostic

Journal of Evaluation in Clinical Practice ISSN 1356-1294

Misleading one detail: a preventable mode of diagnostic

error?jep_1098 804..806 Shahar Arzy MD PhD,1 Mayer Brezis MD MPH,2 Salim Khoury MD,3 Steven R. Simon MD MPH4 and Tamir Ben-Hur MD PhD5

1Neurologist, Department of Neurology, Hadassah Hebrew University Hospital, Jerusalem, Israel and Research fellow, Laboratory of Cognitive Neuroscience, Brain Mind Institute, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland 2Professor, Centre for Clinical Quality and Safety, Hadassah Hebrew University Hospital, Jerusalem, Israel 3Neurologist, Department of Neurology, Hadassah Hebrew University Hospital, Jerusalem, Israel 4Associate Professor, Department of Ambulatory Care and Prevention, Harvard Medical School and Harvard Pilgrim Health Care, Boston, MA, USA 5Head and Professor of Neurology, Department of Neurology, Hadassah Hebrew University Hospital, Jerusalem, Israel

Keywords Abstract clinical decision making, clinical errors, cognitive disposition Rationale, aims and objectives Despite advances in our understanding of cognitive biases in clinical practice, little is known about correction or prevention of diagnostic errors. The Correspondence presence of a single misleading detail may lead clinicians down a cognitive and actual path Dr Shahar Arzy toward an incorrect diagnosis. Department of Neurology Methods In a large teaching hospital, we surveyed 51 attending doctors in internal medi- Hadassah Hebrew University Hospital cine, presenting each with 10 clinical vignettes and soliciting their diagnosis of the con- 91120 Jerusalem dition leading to the presentation. Each of the 10 clinical cases included a single misleading Israel detail. E-mail: [email protected] Results This survey elicited a wrong diagnosis in 90% of cases, which was reduced to 30% when omitting the misleading detail from the vignette. Diagnostic accuracy did not Accepted for publication: 15 July 2008 improve by warning doctors about potentially misleading information. Asking doctors to identify a leading diagnostic detail and then to formulate an alternative diagnosis after doi:10.1111/j.1365-2753.2008.01098.x omission of the detail, signiﬁcantly reduced diagnostic error rate by nearly 50%. Conclusion Systematic re-examination of leading diagnostic clues may help to reduce errors in diagnosis.

Introduction to reduce the number of false MODEs that result in erroneous medical decisions. However, the presence of true MODEs that Errors in diagnosis occur frequently, delaying identification of the divert clinical thinking may be very common. Recognition of both correct diagnosis and resulting in adverse clinical events. Incorrect types of errors might allow reduction of mistakes in clinical deci- or delayed diagnoses are the most common reason for medical sion making. We therefore undertook the present study to investi- malpractice claims in the USA [1]. Despite advances in our under- gate the relationship between MODEs and clinical diagnostic standing of cognitive biases in clinical practice [2–4], little is errors among practicing doctors. known about how overcoming those biases may prevent or correct errors in diagnosis. In clinical thinking, one prominent detail Methods might modify an otherwise simple clinical picture. This misleading one detail (MODE) might originate from false information Fifty-one attending doctors in internal medicine at a university (false MODE: incorrect history or physical examination, misinter- hospital were recruited to participate in this study. Participants pretation or misreporting of imaging or laboratory results), or real were informed that the purpose of the study was to evaluate the findings that are nevertheless not related to the present problem knowledge of internists at the hospital. The 51 participants were (true MODE: e.g. social or demographic characteristics, prior con- randomly allocated to three separate groups: one group received dition or current co-morbidity, or accurate but irrelevant finding on the clinical case with an embedded MODE but received no special physical examination, laboratory investigation or diagnostic notification or instruction to be aware of MODEs (MODE group, imaging). A major goal of clinical quality and safety in medicine is n = 17). A second group received the same clinical cases but was

804 © 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd, Journal of Evaluation in Clinical Practice 15 (2009) 804–806 S. Arzy et al. Preventable mode of diagnostic error instructed to be aware of possible misleading details in case (MODE-notified group, n = 17). A third group received the same case but with a trivial detail substituted for the MODE (non- MODE group, n = 17). Each doctor was asked to read 10 clinical case presentations, composed from five different details each. Cases were based on real cases witnessed in the emergency room, in which a misleading detail caused misdiagnosis (Box 1). For each case, participants were first asked to indicate the most prob- able diagnosis and the first management option to be taken. After completing this step, MODE group and MODE-notified group participants were instructed to indicate the most influential detail in each case. Participants were then asked to reconsider their diagnosis and management approach if the most influential detail they identified had been absent. Each correct response (see Box 1) was credited with one point, and the mean numbers of correct responses per case in each study group were then compared. Repeated-measures anovas were run on the responses with the factor MODE between the three groups. A second statistical test (repeated-measures anovas) was then run with the factors notification (with and without) and revision (before and after) between the MODE group and MODE-notified group. Then, to test possible relationships between MODE’s identification and the revised answers, correlation analysis was performed between responses for the different questions and the corrected answers. Analyses were performed for the whole sample, using Pearson product moment correlations. All P-values are two-tailed, and the signifi- cance level was set to a = 0.05.

Figure 1 Analysis of doctors’ responses. (a) Percentages of correct Box 1 Examples of clinical cases. Each case was comprised from responses are plotted separately for the three different groups (MODE five different details with a misleading one detail (MODE; bolded) in group, MODE-notified group, non-MODE group) before (grey) and after the patient’s history (case a), physical examination (case b) or inves- (black) MODE identification. Note the lower results for the MODE and tigation (case c). The MODEs could be true (cases a,b) or false (case MODE-notified group than that for the non-MODE group as well as the c). The right answer is written in brackets following by the common significant improvement after MODE identification (grey vs. black bars). answer of participants. (b) Revised results for the MODE group are plotted as a function of MODE identification showing a positive correlation. MODE, misleading a. A generally healthy 21-year-old girl is referred for the fourth time one detail. to the emergency room, complaining of pain in her right lower ribs since falling down during skiing 3 months ago. Physical examination reveals a hard thin painful mass above the right lateral lower group and the MODE-notified group (notification effect: aspects of ribs 5–7. Chest X-ray is within normal limits (non- F = 0.23; P = 0.64). The average number of correct diagnoses Hodgkin’s lymphoma/trauma). (1,9) significantly improved in these study groups when participants b. A 67-year-old man, university professor, who is known to suffer were instructed to disregard the most influential detail and revise from multiple vascular risk factors, is brought to the emergency < room after being found in a confused state with difficulty speaking. their decision (revision effect: F(1,9) = 16.4; P 0.01; Fig. 1a). The The nurse reports a blood pressure of 188/96 and the ECG is increase attributable to the instruction to disregard the MODE was within normal limits (hypoglycaemia/stroke). similar in both groups, as no significant interaction was found c. A 40-year-old woman is suffering from general weakness and between effects (F(1,9) = 0.004; P = 0.95). Post hoc (Scheffé) shortness of breath. Two weeks ago she was hit on her chest while analysis showed significant improvement after revision in both sitting on a ski elevator. Physical examination reveals reduced ven- MODE and MODE-notified groups (P < 0.001 for both groups). tilation over the lower left lung and chest X-ray shows left lower Notably, MODE-identification correlated positively with the lobe infiltrate (haemothorax/pneumonia). revised responses in the MODE group (r = 0.96; P < 0.001; Fig. 1b) as well as in the MODE-notified group (r = 0.91; P < 0.001). Results Discussion The average number of correct initial diagnoses in the MODE group and MODE-notified group was significantly lower than A survey of physicians in internal medicine using clinical vignettes the non-MODE group (MODE effect: F(2,18) = 81.5; P < 0.001; showed that a single misleading detail (MODE) can elicit a wrong Fig. 1a). There was no significant difference between the MODE diagnosis in 90% of cases, reducing to 30% when omitting the

© 2009 The Authors. Journal compilation © 2009 Blackwell Publishing Ltd 805 Preventable mode of diagnostic error S. Arzy et al. detail from the vignette. Diagnostic accuracy did not improve by Acknowledgement warning doctors about potentially misleading information. These results show that a MODE, true or false, originated at any step of S. A. is supported by the Yeshaya Horowitz Association through the clinical investigation, might drastically bias diagnosis and the Center of Complexity Science. treatment. Indeed, biases and dispositions in rational thinking are already recognized for more than 30 years [5]. Many authors References discussing biases in the context of medical decision making – such as the first impression (‘anchoring bias’), previous experience 1. Groopman, J. (2007) How Doctors Think. Boston, MA: Houghton (‘availability’) or prejudice (‘premature closure’) – conclude that Mifflin Company. awareness of these biases are important in improving clinical 2. Croskerry, P. (2000) The cognitive imperative: thinking about how we think. Academic Emergency Medicine, 7 (11), 1223–1231. results [2–4,6,7]. Nevertheless, as was also shown here, mere 3. Croskerry, P. (2002) Achieving quality in clinical decision making: awareness to these biases does not significantly reduce their influ- cognitive strategies and detection of bias. Academic Emergency Medi- ence [5,8]. By contrast, asking doctors to identify a leading diag- cine, 9 (11), 1184–1204. nostic detail and then to formulate an alternative diagnosis after 4. Voytovich, A. E., Rippey, R. M. & Suffredini, A. (1985) Premature omission of the detail significantly reduced diagnostic error rate by conclusions in diagnostic reasoning. Journal of Medical Education,60 nearly 50%. While many heuristics-based clinical decisions are (4), 302–307. quickly and correctly made, they might be nevertheless interfered 5. Tversky, A. & Kahneman, D. (1974) Judgement under uncertainty: by factors such as a MODE. The present study, performed on heuristics and biases. Science, 186, 1124–1131. vignettes, cannot predict whether doctors would be able in practice 6. Hall, K. H. (2002) Reviewing intuitive decision-making and uncer- to reduce diagnostic errors by conscientious omission a leading tainty: the implications for medical education. Medical Education,36 (3), 216–224. detail in appropriate cases – without increasing errors in other 7. Kassirer, J. P. & Kopelman, R. I. (1989) Cognitive errors in diagnosis: corrected diagnosed patients. However, in cases which are not instantiation, classification, and consequences. The American Journal straight forward, such as repeated referral to medical assistance, of Medicine, 86 (4), 433–441. patients with preliminary disease or risk factors or unanticipated 8. Graber, M. L., Franklin, N. & Gordon, R. (2005) Diagnostic error in complications, systematic re-examination and verification of internal medicine. Archives of Internal Medicine, 165 (13), 1493– leading diagnostic clues may help to reduce diagnostic mistakes. 1499.