A 54-year-old homeless man with a history of alcoholism presented to the emergency department (ED) with complaints of knee problems. The triage nurse documented the chief complaint as "bilateral knee pain" and left the chart for the ED physician. The patient had not experienced any trauma to the knees and had no other symptoms; a focused physical examination of the knees was unremarkable. The ED physician diagnosed the patient with a musculoskeletal injury and prepared to discharge him. After receiving his discharge instructions, the patient tried to get up and walk but was noted to be unsteady. A subsequent full neurologic examination raised additional concerns and a diagnostic head computed tomography (CT) showed a subdural hematoma. The patient was admitted for urgent neurosurgical intervention.
Given the near miss of an unsafe discharge and the initial diagnostic error, the hospital's quality committee formally reviewed the case. The triage nurse reported that the patient used a number of vague and varied complaints about his knees, such as: "giving out," "couldn't walk on them," and "feeling wobbly." The nurse simply summarized the descriptions with the term "bilateral knee pain." The ED physician relied heavily on this documented triage complaint, leading to an overly focused history and examination.
Diagnostic errors are estimated to occur in up to 15% of cases (1) and are an important frontier in health care quality and safety. Although this patient was initially misdiagnosed, formal analysis of the case requires a shared understanding of where in the diagnostic spectrum the error occurred and where opportunities for improvement lie. Following Schiff and Leape's taxonomy (2), three questions must be addressed:
Was there a diagnostic process error?
Yes—the initial history and physical exam were mistargeted.
Was there a diagnostic error (i.e., missed or delayed)?
Yes—the diagnosis of subdural hematoma was delayed.
Was there an adverse outcome?
No—during the same episode of care, the patient was treated for the life-threatening emergency without suggestion that the delay led to impaired neurologic outcome. But this was a near miss.
Triage and Diagnosis
Triage is not designed to be part of the diagnostic process, which traditionally begins when the provider asks the patient "How can I help you?" The goals of triage are to promptly identify patients requiring immediate treatment of life-threatening conditions and to prioritize the care of all other patients based on their acuity and available resources.(3) However, the cognitive and social reality is that triage often plays a de facto role in the diagnostic process. This case exemplifies how multiple latent factors embedded in triage can initiate a cascade of events leading to a diagnostic error.
The cognitive demands of triage are substantial. In a brief encounter, the triage nurse compiles and analyzes information from a brief history, visual inspection, and vital signs, and is asked to concisely convey the patient's condition and chief complaint to a treating clinician who may be separated in space and time from the nurse. Patients may provide inaccurate information on account of privacy concerns, confusion, wanting to tell story only once, or knowledge of "trigger words" that expedite care (e.g., a patient with "full body pain" may stress chest pain to the triage nurse—understanding that this may prioritize care given the many life-threatening diagnoses, like myocardial infarction or pulmonary embolism, associated with this complaint). The time limitations of triage make detailed investigations into these matters extremely challenging and oftentimes impractical. Given the structure of the encounter, it is not reasonable to routinely expect a sound diagnosis—or sometimes even a crystallized chief complaint—to reliably emerge. Here, the triage nurse's summary of "bilateral knee pain" was not synonymous with the varied issues heard from the patient. However, this chief complaint should not have constrained the physician.
The human mind's initial impressions and final decisions are strongly influenced by the way information is presented (framing effect). Decision makers have a related tendency to rely heavily on isolated pieces of information that reinforce initial hypotheses (anchoring effect). The ED physician receives information framed from multiple sources, including calls from referring hospitals or clinics with initial impressions, opinions from the patient, and triage data. The possibility that this ED physician relied too heavily on the triage chief complaint invites opportunities to systematically address this cognitive pitfall. Simply pointing out that a predisposition to think in a certain way exists, referred to as a heuristic, will not stop a physician from being susceptible to it. Rather, counterforcing strategies (4) must be engineered to trigger caution when hard-wired and familiar cognitive processes are at play. For instance, a computer program or checklist could automatically prompt a clinician to ask "Have I ruled out the worst case scenario?" or "Have I considered an alternative diagnosis?" before treatment for the working diagnosis is administered. It must be acknowledged, however, that a counterforcing function is designed to override a cognitive process that generally works well and thereby runs the risk of inducing excess costs and harms.(5,6)
One approach is to incorporate a trigger in the handoff communication between the triage and treating providers. Systems of communication can reinforce the provisional nature of the initial triage assessment by incorporating fields entitled "TENTATIVE assessment" or "INITIAL chief complaint." This approach would require revision of spoken terminology or modification of handwritten or computerized tools, but has the potential to reduce the weight given to early reports and introduce a healthy skepticism that mitigates framing bias.
Because high reliability systems depend on redundancy to prevent errors, additional workplace tools should be considered to increase the fidelity of the diagnostic process. Checklists are used in error-prone and complicated processes. Ely and colleagues (7) have proposed the use of a diagnostic checklist designed to aid and offload clinicians' intuition and memory when solving complex clinical problems. The first step of their proposed General Checklist for Diagnosis is to "obtain your own complete medical history," which serves as another trigger to revisit preliminary information and avoid the framing bias. A sub-step could be added to specifically address the issue in this case, such as "I have revisited the referring/triage complaint and confirmed or revised it."
Plenty to Consider
Systems to improve diagnosis are in their infancy.(8) As they improve and become more widespread, there is risk that physicians will reject them, fearing "cookbook" medicine or being unfaithful to the defining cognitive process of the profession. This fear is misplaced, as tools currently in development—communication strategies, forcing functions, diagnostic checklists, and decision support—serve as diagnostic adjuncts and rely heavily on physician cognition analysis before and after their deployment. None pretend to have an "autopilot" function that excludes the clinician.
Diagnostic tools do move the physician's cognitive challenges upstream to detect, identify, and frame problems (i.e., enter the correct issue) and downstream to analyze and apply the results or prompts (e.g., stop to consider alternative diagnoses or decide which newly suggested diagnoses are worth excluding). In between, these tools are meant to offload tasks that are straightforward but forgettable (e.g., start over by taking your own history), allowing the brain to concentrate on higher-level thinking and increased vigilance (e.g., possibly detecting other signs of subdural hematoma). None of the tools will come with a guarantee of providing the correct diagnosis, but the use of multiple embedded approaches can increase those odds. For instance, it is hard to imagine how any computerized diagnostic algorithm could begin with knee pain and conclude with subdural hematoma. Conversely, steadfast adherence to a diagnostic checklist would have likely steered the clinician in the right direction.
Diagnosis is a complicated and error-prone process. Physicians should rest assured that diagnosis will rely heavily on their own reasoning for the foreseeable future. But they should recognize the limits of human cognition and memory and welcome any interpersonal, organizational, or technological efforts to make the process as safe, accurate, and reliable as possible.
- Diagnostic process errors, diagnostic errors, and adverse outcomes have overlapping but distinct definitions.
- Multiple latent factors embedded in triage can initiate a cascade of events leading to a diagnostic error.
- Communication tools and diagnostic checklists can be integrated into workflows to mitigate errors induced by the framing effect (the tendency to accept information as presented).
- Current and proposed diagnostic support tools aim to increase accuracy and safety but pose little threat of "autopilot" or "cookbook" medicine.
Krishan Soni, MD, MBA Chief Resident for Quality Improvement and Patient Safety Department of Medicine University of California, San Francisco
Gurpreet Dhaliwal, MD Associate Professor of Clinical Medicine Department of Medicine University of California, San Francisco San Francisco VA Medical Center
1. Shojania KG, Burton EC, McDonald KM, Goldman L. Changes in rates of autopsy-detected diagnostic errors over time: a systematic review. JAMA. 2003;289:2849-2856. [go to PubMed]
2. Schiff GD, Leape LL. Commentary: how can we make diagnosis safer? Acad Med. 2012;87:135-138. [go to PubMed]
3. Washington DL. Triage time bomb. AHRQ WebM&M [serial online]. January 2004. [Available at]
4. Croskerry P. Cognitive forcing strategies in clinical decisionmaking. Ann Emerg Med. 2003;41:110-120. [go to PubMed]
5. Sherbino J, Yip S, Dore KL, Siu E, Norman GR. The effectiveness of cognitive forcing strategies to decrease diagnostic error: an exploratory study. Teach Learn Med. 2011;23:78-84. [go to PubMed]
6. Norman G. Dual processing and diagnostic errors. Adv Health Sci Educ Theory Pract. 2009;14(suppl 1):37-49. [go to PubMed]
7. Ely JW, Graber ML, Croskerry P. Checklists to reduce diagnostic errors. Acad Med. 2011;86:307-313. [go to PubMed]
8. Umscheid CA, Hanson CW. A follow-up report card on computer-assisted diagnosis—the grade: C+. J Gen Intern Med. 2012;27:142-144. [go to PubMed]