How to Build a Differential Diagnosis: A Step-by-Step Framework

Picture this: a 58-year-old man walks into the emergency department with chest pain. He's diaphoretic, his blood pressure is 90/60, and he's been having the pain for 45 minutes. What's your differential?

If your brain immediately went to "MI" — that's pattern recognition, not differential diagnosis. Pattern recognition is powerful and usually right. But it's also how diagnostic errors happen. The best clinicians do something different: they build a differential systematically, then use pattern recognition to rank it.

Here's the framework.

Step 1: Anchor on the Presenting Problem, Not the Diagnosis

The first rule of differential diagnosis: start with what the patient is telling you, not what you think they have. The presenting problem is chest pain, not "STEMI." This sounds obvious, but it's where most students go wrong — they anchor on the most dramatic or memorable diagnosis and work backwards.

Before generating your differential, define the presenting problem precisely:

• What is the primary symptom? (chest pain, dyspnea, altered mental status)
• What are its key characteristics? (onset, duration, quality, location, radiation)
• What makes it better or worse?
• What are the associated symptoms?

Step 2: Generate by Anatomic or Pathophysiologic Category

Don't try to recall a memorized list. Instead, generate your differential systematically by asking: "What structures are in this region, and what can go wrong with them?"

For chest pain, work through the anatomy of the chest:

• Heart: ACS, pericarditis, myocarditis, cardiac tamponade
• Aorta: Aortic dissection
• Lungs/pleura: PE, pneumothorax, pneumonia, pleuritis
• Esophagus/GI: Esophageal spasm, GERD, peptic ulcer perforation
• Chest wall/musculoskeletal: Costochondritis, rib fracture, herpes zoster
• Referred pain: Subphrenic abscess, hepatobiliary disease
• Psychiatric: Panic disorder, somatic symptom disorder

This approach ensures you don't miss an entire category. It also forces you to think about pathophysiology — why would each structure cause this symptom? — which deepens your understanding far beyond memorization.

Step 3: Apply the Probability Framework

Not all differentials are created equal. Once you've generated a broad list, rank it using three categories:

Step 4: Use the Clinical Data to Narrow

A differential is a living document. As you gather more data — physical exam findings, vital signs, initial labs, ECG — you update your probability estimates and narrow your list.

Think of each piece of clinical data as a likelihood ratio. The ECG showing ST elevations dramatically increases the probability of STEMI and dramatically decreases the probability of esophageal spasm. The physical exam finding of unequal blood pressures in both arms dramatically increases the probability of aortic dissection.

Step 5: Commit, But Stay Open

The final step — and the hardest — is committing to a working diagnosis while maintaining intellectual humility. You need a working diagnosis to drive management. But you also need to stay open to revision when the patient doesn't respond as expected or new data emerges.

The biggest diagnostic errors happen when clinicians commit too early and stop updating their differential. In the cognitive bias literature, this is called premature closure — anchoring on the first plausible explanation and failing to consider alternatives.

The antidote is a simple habit: every time you see a patient back, ask yourself — "Is there anything about this presentation that doesn't fit my working diagnosis?" If yes, reopen the differential.

Putting It Together: The One-Minute Differential

In practice, experienced clinicians do all five steps in under a minute. Here's how to develop that fluency:

1. State the presenting problem precisely — "acute-onset tearing chest pain radiating to the back in a hypertensive 58-year-old man"
2. Name your anatomic categories — cardiac, vascular, pulmonary, GI, musculoskeletal
3. Generate 1-2 conditions per category — don't try to be exhaustive at this stage
4. Apply the probability framework — most likely, must not miss, reasonable
5. Identify the one test that will most change your management — this drives your immediate workup

This isn't a formula to follow rigidly — it's a scaffold to internalize. The goal is for it to become automatic, so that even under pressure, your reasoning stays systematic rather than reactive.

How to Practice

Clinical reasoning is a skill, not knowledge. You improve it the same way you improve any other skill: deliberate practice with feedback. Reading about differential diagnosis helps. Generating differentials on real or simulated patients, then getting coaching on your reasoning — that's what actually builds the skill.

This is why we built ReasonDx. Not another question bank, not another set of notes to read — a platform where you practice the actual cognitive process of clinical reasoning with an AI coach who asks you the right questions at the right moments.