CardioExchange Archive

A 55-year-old man came to the emergency room complaining of aching chest pain radiating to the back. The pain had started the day before and recurred several times. It seemed to worsen with exertion and resolve with rest. One resting episode was associated with diaphoresis. Exam, EKG, and cardiac enzymes were normal. A portable chest X-ray showed a slightly widened mediastinum. The ER resident ordered a CT scan of the chest, and the preliminary report was negative for aortic dissection but showed a small pericardial effusion. The patient was placed on aspirin, a beta-blocker, and heparin. A cardiology consultation was called, for presumed unstable angina.

I was on call and saw the patient that Saturday afternoon. After talking with him, I began to question the preliminary CT scan report. I asked an attending radiologist to go over the scan with me, and on careful review, a subtle dissection was noted. The heparin was stopped, and the patient underwent emergency aortic repair. The ER resident was stunned.

“What caused you to doubt the preliminary CT scan report?” he asked. “How did you know?”

“The scan result just didn’t fit,” I said.

Had I simply made a lucky guess? Perhaps. But after eliciting a history, examining the patient, and going over the other details, I formed a mental picture of the case that made me question the preliminary CT scan report.

How do experienced doctors “see” a diagnosis and start to solve a diagnostic problem? How do we begin to generate the “early hypotheses” that I discussed in my previous post? The answer, in my view, lies in pattern recognition.

The Way We Detect Patterns

During the first two years of medical school, we learn about causal reasoning — how basic biologic defects manifest in disease. Unfortunately, patients don’t come to us with an orderly textbook description of a condition. They present with signs and symptoms, forcing us to work backward in seeking a diagnosis. The philosopher Charles S. Peirce described this abductive reasoning as “regressing from a consequent to a hypothetical antecedent.” To begin to work backward, we have to recognize a pattern. The pattern may not be clear initially, or it may lack some key parts — but eventually one does emerge, and it starts us on the path of diagnostic evaluation.

Pattern recognition reinforces two traditional methods in medical practice: the use of narrative and the differential diagnosis. Kathryn Montgomery discusses the use of narrative in her book How Doctors Think. While in training, we physicians present cases repeatedly, practicing the art of developing stories that have meaning. The narratives are detailed, chronological, and structured by events. As we develop a narrative, we set ourselves up to detect patterns.

Judith Bowen has discussed how we transform the details of a patient’s case into a mental abstraction called a “problem representation” (N Engl J Med 2006; 355:2217). I like to think of this transformative step as being similar to solving an algebraic equation — sorting, factoring, canceling, and simplifying. The process starts with active listening, and it requires attentiveness and cognitive energy. We then compare the problem representation with “illness scripts” that are stored in our memory from past clinical decision-making experiences. When we find a match, we have a hypothesis or plausible conjecture that we test (usually against imaging or blood-test results) to work toward a final diagnosis.

But this process has its pitfalls. One is called post hoc, ergo propter hoc — a Latin phrase that describes the faulty assumption that temporal sequence implies causation. Another trap is confirmation bias, or selective use of facts to conform to a desired story. A third trap is anchoring, or placing too much weight on a particular aspect of a story. As we draw inferences from a particular case to arrive at a general principle or conclusion (i.e., as we reason abductively), we must of course avoid hasty generalizations or overgeneralizations. That’s where our training in differential diagnosis comes into play — it helps us entertain alternative diagnoses so that we don’t leap to erroneous conclusions.

Reflecting on how we think, how we generate hypotheses, and how we test them makes us more mindful about the art of arriving at a final diagnosis. In this era of imaging and biomarkers, do we need these diagnostic reasoning skills as much as we once did? To what extent has the art of diagnostic reasoning changed because of our new tools and tests?