A 67-year-old man presents to the hospital with persistent, subjective fevers and malaise for one month, subacute onset of dyspnea, and nonproductive cough for the preceding six days. The patient is a nonsmoker, denies sick contacts, and has had no foreign travel. What would be the best approach to making the diagnosis while working to enhance diagnostic skills?
With clinical experience, making a diagnosis can become so routine that physicians might not contemplate their problem-solving strategies. Diagnostic reasoning is the process of thinking about a clinical problem to form a diagnosis. Experienced clinicians typically rely upon nonanalytic reasoning (i.e., pattern recognition) for straightforward problems, reverting to analytic reasoning if a pattern is not recognized.
The literature describes five steps in the reasoning process (see Figure 1). In the early stages of data collection, hypotheses emerge that feed back into data collection behaviors as the clinician seeks confirmatory evidence. This complex interplay between data collection and hypothesis generation/elimination leads to a more clearly defined understanding of the patient’s presentation. The synthesis of the patient’s presentation, including epidemiologic risk factors, symptoms, signs, and laboratory and radiologic studies, is called the “problem representation.” After a clinician conceives the problem representation, he or she reviews the mental representations of diseases (i.e., illness scripts) to determine hypotheses by finding disease presentations that best match the formulated problem representation (see Figure 2).
Analytic and nonanalytic reasoning. In what is known as the dual process theory, diagnostic reasoning is believed to occur both analytically and nonanalytically.1 Nonanalytic reasoning is often exemplified by rapid, subconscious “pattern recognition” and is developed through clinical experience and other nonclinical learning experiences (e.g. reading).
Conversely, analytic reasoning, the “slow,” conscious, cognitive processing, is typically utilized when a patient presentation is complicated or does not fit a known disease pattern. Clinicians apply both strategies to make diagnoses in evaluating complex cases.
In the outlined case, while the symptoms of fever and cough might lead to the diagnosis of community-acquired pneumonia (CAP), the time course seems unusually long. This atypical pattern for CAP could trigger analytic reasoning, leading to new considerations such as tuberculosis (TB).
On examination, the patient has severe rigors and diaphoresis, as well as a fever of 39.4°C and a heart rate of 102 bpm. Full examination discloses mild end-expiratory wheezes and bronchial breath sounds in the right lower lobe. The remainder of his examination is normal. Labs reveal WBC 8.5x103, hemoglobin 11g/dL, MCV of 92 fL, and platelet count 22,000 mm3. Blood cultures, sputum cultures, and respiratory virus microarray are normal. The chest X-ray (CXR) is unremarkable.
Further history reveals that the patient is a sheepherder living in a primitive earthen structure in the rural mountains of western New Mexico.
Problem representation revisited. With additional historical, laboratory, and radiological data collected, further interpretation and synthesis occur. Salient elements are highlighted and prioritized, irrelevant details are discarded, and data of uncertain relevance are reevaluated as additional data are gathered. The problem representation—an interpreted, subjective mental model of a patient’s clinical presentation—is updated and reformulated. The verbal expression of the problem representation is variously called the assessment, summary statement, or “one-liner.” Within this summary statement, and fundamental to the creation of a strong problem representation, is the incorporation of “semantic qualifiers.”
“Semantic qualifiers” (e.g. acute vs. chronic or unremitting vs. relapsing) are paired, opposing descriptive adjectives that can be used to compare and contrast diagnostic considerations.² Clinicians distinguish between diseases using key signs and symptoms and use these descriptors to assist with this discrimination in hypothesis generation. An example for this patient would be: A 67-year-old sheepherder living in rural New Mexico presents with persistent fevers and malaise for one month, along with subacute development of nonproductive cough and dyspnea, sepsis, anemia, and thrombocytopenia.
Note how the incorporation of epidemiologic information (sheepherder living in an earthen structure in rural New Mexico) creates a context in which the additional problems can be framed (persistent malaise, subacute cough). In this case, the persistent fevers help the clinician to narrow possibilities in the differential diagnosis and create focused hypotheses.
Although the benefit of teaching accurate and thorough problem representation seems self-evident, studies have not demonstrated that improved problem representation enhances diagnostic accuracy; however, we believe that there is still value in adapting and teaching this skill.3
Hypothesis refinement and the differential diagnosis. Initial hypotheses occur early in data collection, as the patient’s history and physical examination findings trigger connections to clinicians’ bank of known diseases (e.g. orthopnea triggers congestive heart failure). As the clinician collects additional data, he or she refines these hypotheses, changing the likelihood based on “fit” of the problem representation with known diseases or illness scripts.
When employing analytic reasoning processes, clinicians may benefit from using organizational frameworks to assist with hypothesis generation (see Table 1). For this patient, possible hypotheses could include CAP, TB, lymphoma, lung neoplasm, or other indolent pulmonary infection.
Illness scripts. Once discrete hypotheses (e.g. CAP, pulmonary embolism) have been generated, clinicians need a method to accurately compare disease processes. This can be done through the use of an illness script. Illness scripts are mental representations of diseases and are likely to include epidemiology, typical and atypical patterns of presentation, and distinguishing features.
For example, a clinician’s illness script for a typical presentation of bacterial CAP likely includes fever, productive cough, pleuritic chest discomfort, and infiltrate on CXR. Clinician educators who teach illness scripts should ensure that students understand that diseases have atypical presentations, even though they may only teach them the prototypical one. Conceptualizing diseases in this fashion allows clinicians to seek the disease with the “script” that best matches the patient’s story (i.e., clinical presentation).
In this case, the clinician is now thinking of causes of persistent fever + nonproductive cough + dyspnea + anemia + thrombocytopenia; possibilities include lymphoma or unusual infection (e.g. tick-borne relapsing fever, or TBRF).
Case Resolution and Script Selection
As the clinician processes the case, a known illness script of TBRF matches the patient’s clinical presentation, and a peripheral smear is ordered. The smear reveals presence of spirochetal organisms, later confirmed by PCR to be Borrelia hermsii, confirming the diagnosis of TBRF.
Errors in Clinical Reasoning
Although most clinicians are quite accurate in typical presentations of common diseases, they are more likely to commit diagnostic errors when faced with uncommon diseases, atypical presentations, and/or challenging contexts. The following sections categorize a selection of some common errors and offer some expert opinion from the literature on avoiding them.
Common diagnostic errors. Clinicians use heuristics, or mental shortcuts, which can occasionally induce diagnostic errors. By definition, the fundamental problem in all diagnostic error is premature closure, or acceptance of a diagnosis before it is fully verified. In the case presented, the clinician may have accepted the diagnosis of CAP without recognizing other possible diagnoses.
Two common heuristics/biases that can sometimes lead to premature closure are the availability and anchoring biases. Availability bias means that the diagnoses easily thought of—and often most recent in the memory—are more likely to be assigned to a patient problem. The diagnosis of pulmonary embolism would be more “available” in a patient with fever, dyspnea, and normal CXR, especially if the clinician recently had seen a patient with PE. Anchoring bias occurs when early information is relied upon to make clinical judgments and the clinician fixates on a diagnosis despite acquiring additional or contrary information. For example, a clinician may rely upon a diagnosis of CAP based on the sign-out from a colleague, despite the one-month history of symptoms, rather than broadening the differential.
Clinician-focused methods to reduce diagnostic errors. Multiple methods exist that may mitigate diagnostic errors, although definitive proof of their value is still lacking, owing to the difficulty involved in studying such errors due to the multitude of causes.4 In our opinion, building a mental database of illness scripts by reading and seeing patients, as well as being metacognitive, are the best methods for individual clinicians to use to reduce their errors (see “deliberate practice” below).
Metacognition, or thinking about one’s thinking, is another method of reducing errors and can be characterized by “reflection in action” (reflection in real time) and “reflection on action” (reflection after an event).5 For example, taking a few moments at the end of a week on clinical service to reflect on the hospital course and diagnostic paths of the most complex patient presentations (reflection on action) is an exercise used to reduce errors.
For reflection in action, a clinician may pause when confronted with paradoxical findings for a current patient’s presentation (e.g. elevated jugular veinous pressure and crackles on exam but normal b-type natriuretic peptide), and “think aloud” (see below) to ensure he or she is processing all of the appropriate elements of the case.5
In the case presented above, the time course might have initiated reflection into erroneous decision-making at the moment the clinician thought that CAP was a possibility (reflection in action). Although direct evidence is inconclusive as to whether these techniques improve diagnostic accuracy, engaging in metacognitive exercises remains a cornerstone of seasoned clinical reasoning experts.6
Teaching and Learning Principles
Making a commitment. During a patient presentation, it is often helpful to ask a learner to develop a two- to four-item prioritized differential diagnosis list based on likelihood and/or lethality. Have the learner describe which diagnosis is most likely (i.e., the working diagnosis), in addition to the reasons “for” or “against” certain hypotheses. Once the diagnosis has been determined, combine commitment with an exercise in metacognition by asking the learner, “Why do you think that your initial diagnosis of Q-fever was incorrect?” Clinical educators may then follow up with teaching pearls and their approach to this type of case (see Table 1).
Think aloud: In this method, an instructor expresses his or her thoughts aloud in real time.7 By modeling this technique, attending physicians allow learners to observe the process of developing a differential diagnosis and plan. For example, during the admission process, instructors could verbalize their approach to fever in a systematic fashion (see Table 1) after the trainee has completed the presentation: “At this point I am considering an infectious cause such as pneumonia, given the respiratory symptoms, although the one-month history of fever and malaise makes me think that I should keep neoplasm and an unusual infection on my list of possibilities.”
Conversely, instructors can ask trainees to voice their thoughts aloud to better understand their reasoning processes. By using this method, instructors can also support, correct, or reinforce the trainees’ appropriate use of knowledge in the clinical reasoning process.
Deliberate practice. To improve diagnostic skills, trainees must engage in deliberate practice, defined as intentional, repetitious practice aimed at improving performance.8 To facilitate this, a trainee should evaluate as many patients as possible and present to an experienced clinician with subsequent feedback. Trainees are likely to miss subtle historical or examination points (e.g. the history of sheepherding) because their illness scripts are limited or incompletely developed. Teachers should emphasize the importance of developing broad and deep illness scripts, so learners will, hopefully, become more aware of their limitations and recognize what they do not know.
Clinicians solve diagnostic problems using both nonanalytic and analytic reasoning processes. Although evidence is inconclusive, some clinical reasoning experts suggest the use of reflective strategies to enhance diagnostic accuracy, especially in complicated cases.9 To prevent premature closure, we encourage hospitalists to perform an analytic “double-check” before determining their final diagnosis.
Furthermore, the clinical reasoning literature suggests that knowledge and its organization are key to expert performance.10 In diagnostic reasoning, this key knowledge has been termed “illness scripts.” Thus, the task of the aspiring expert diagnostician is to learn the key features of diseases and focus on discriminating features, starting with typical presentations of common diseases and working up to atypical presentations of uncommon diseases.
Engaging in deliberate practice, seeking feedback on diagnostic accuracy, and reflecting upon your own reasoning process can provide valuable information for improving future diagnostic reasoning. The ultimate goal of these practices is to enhance diagnostic skills in order to avoid errors and improve patient care.
Diagnosis is a challenging task. Diagnostic accuracy may be enhanced by expanding the learner’s knowledge of illness scripts and using an analytic double-check to confirm initial diagnoses determined by nonanalytic reasoning.
Drs. Rendon, Roesch, and Rao are hospitalists in the department of internal medicine at the University of New Mexico School of Medicine in Albuquerque. Dr. Rencic is a hospitalist in the department of internal medicine at Tufts University School of Medicine in Boston.
- Eva KW. What every teacher needs to know about clinical reasoning. Med Educ. 2005;39(1):98-106.
- Bowen JL. Educational strategies to promote clinical diagnostic reasoning. N Engl J Med. 2006;355(21):2217-2225.
- Nendaz MR, Bordage G. Promoting diagnostic problem representation. Med Educ. 2002;36(8):760-766.
- Norman GR, Eva KW. Diagnostic error and clinical reasoning. Med Educ. 2010;44(1):94-100.
- Schön DA. The Reflective Practitioner: How Professionals Think in Action. London: Temple Smith; 1983.
- Croskerry P. Achieving quality in clinical decision making: cognitive strategies and detection of bias. Acad Emerg Med. 2002;9(11):1184-1204.
- Van Someren MW, Burnard YF, Sandberg JAC. The Think Aloud Method: A Practical Guide to Modelling Cognitive Processes. London: Academic Press; 1994.
- Ericsson KA, Krampe RT, Tesch-Romer C. The role of deliberate practice in the acquisition of expert performance. Psychol Rev. 1993;100(3):363-406.
- Mamede S, Schmidt HG, Penaforte JC. Effects of reflective practice on the accuracy of medical diagnoses. Med Educ. 2008;42(5):468-475.
- Elstein, AS, Shulman LS, Sprafka SA. Medical Problem Solving: An Analysis of Clinical Reasoning. Cambridge, Mass.: Harvard University Press; 1978.
- Kassirer JP. Teaching clinical reasoning: case-based and coached. Acad Med. 2010;85(7):1118-1124.
- Rencic J. Twelve tips for teaching expertise in clinical reasoning. Med Teach. 2011;33(11):887-892.