Differential Diagnosis & Clinical Reasoning: How Clinicians Actually Think

The two systems behind every diagnosis

Watch a seasoned clinician see a patient and you’ll notice something odd: they often know the diagnosis in under a minute, but when you ask them how, they struggle to explain. They’ll say something like “He just looked like a PE to me.” That’s not magic. It’s pattern recognition — what cognitive psychologists call System 1 thinking: fast, almost unconscious, built from years of seeing the same patterns. It’s how 80% of routine cases get diagnosed.

The other 20% — the atypical, the rare, the dangerous, the elderly patient with three things going on at once — requires System 2 thinking: deliberate, analytic, slower, and the kind of reasoning that doesn’t fit on a flowchart. Experts don’t just use System 1 better; they have a much better sense of when to drop out of it. That switch is the single most important habit in clinical reasoning.

If a case “feels off,” that feeling is information. Slow down. Ask “what else could this be?” The diagnostic errors that hurt patients almost always involve sticking with System 1 in a case that needed System 2.

Illness scripts — the unit of clinical thought

An illness script is a compact mental model of a disease in three layers:

1. Predisposing conditions— who gets it? (Age, sex, comorbidities, exposures.)
2. Pathophysiology & pace— how does it unfold? Sudden? Days? Weeks?
3. Clinical features & refuters— what does it usually look like, and what would make it less likely?

When an experienced clinician hears “58-year-old male, smoker, central crushing chest pain radiating to the left arm, 30 minutes,” they’re not pattern-matching to a textbook — they’re lighting up an illness script labeled “acute coronary syndrome” that already contains expected ECG findings, expected troponins, what would refute it (the pain that lasted 10 seconds and went away with a sip of water), and the next step.

That’s the structure you want to build for each common disease. Not memorise pages of bullet points — build a small mental story you can tell about each illness. Start with the 30 conditions you’re most likely to see this year and build a script for each.

How to actually build a differential

A common student trap: huge differential, no ranking. “Causes of chest pain include MI, PE, pneumothorax, aortic dissection, pericarditis, GORD, musculoskeletal, panic disorder…” All true. Useless.

A working differential has structure:

• 3–5 most likely diagnoses, ranked by probability given this patient (not a textbook patient).
• 1–2 can’t-miss diagnoses that must be excluded even if less likely, because missing them is catastrophic.
• What you’d need next to distinguish between the top three.

That’s usually about five items total. Anything longer is a brain-dump, not reasoning. Anything shorter probably means premature closure (more on that later).

Prior probability: the thing most students skip

“When you hear hoofbeats, think horses, not zebras.” The cliché is right. Most patients in most settings have common diseases, not rare ones. The probability of a diagnosis depends massively on:

• Setting— a 22-year-old in a student clinic with chest pain is overwhelmingly likely to have musculoskeletal pain or anxiety. The same complaint in an ED has a different prior.
• Patient factors— age, sex, comorbidities, exposures, medications. A 30-year-old non-smoker with no risk factors and pleuritic chest pain has a very different prior than a 65-year-old diabetic.
• Local epidemiology— cough in a TB-prevalent region looks different from cough in a low-prevalence one.

The job is to start with the realistic priors and only move toward the zebra when the evidence specifically demands it. Saying “could be pheochromocytoma” in a hypertensive 60-year-old with no episodic symptoms is not impressive thinking — it’s the absence of it.

“Most likely” vs. “can’t miss”

One of the most useful distinctions in all of clinical reasoning. Every differential has two columns:

• Most likely— high prior probability. Where the workup should mostly point.
• Can’t miss — lower probability, but missing it kills the patient. Must be screened for or excluded.

For chest pain: most likely might be musculoskeletal or reflux; can’t miss is ACS, PE, aortic dissection. The investigations you order on a Tuesday afternoon are shaped as much by the can’t-miss list as by the most-likely one. Senior clinicians screen for the can’t-miss diagnoses every single time, even when they’re sure it’s the boring thing. That habit is patient safety in slow motion.

Narrowing the list with one extra question

The art of clinical reasoning at the bedside is asking the question that splits the differential. If you have ACS, PE, and aortic dissection on the table, the “tearing pain radiating to the back” question separates one. The “recent long-haul flight or immobility” question separates another.

A useful trick to practise: after your initial differential, for the top three, write down one specific history feature, one exam finding, and one investigation that would most change your probability for each. That mental discipline is what stops you from ordering everything and reading nothing.

The cognitive biases that quietly miss diagnoses

Most diagnostic errors aren’t from missing knowledge — they’re from flawed reasoning. The big five worth recognising in yourself:

• Anchoring— locking onto the first diagnosis suggested (often by the referrer or triage note) and weighing later information to fit it. If the referral letter says “query UTI,” you may stop looking for the more dangerous cause of confusion.
• Premature closure— deciding on a diagnosis before you’ve genuinely considered alternatives. Usually a symptom of System 1 running unchecked.
• Availability bias— overweighting diseases you recently saw or recently read about. The week after a near-miss aortic dissection on the ward, everyone’s got chest pain that “could be dissection.”
• Confirmation bias— seeking information that supports your working diagnosis while quietly discounting evidence against it.
• Diagnostic momentum— a label given by an earlier clinician propagates through the chart. You inherit it without re-examining whether it’s right. Especially toxic in elderly patients with vague diagnoses like “chronic confusion.”

You can’t eliminate these biases. But you can build habits that interrupt them: ask “what else could this be?”, take a few seconds to consider the opposite of your current hypothesis, re-examine inherited diagnoses on day three of admission. Small habits, huge yield.

Metacognition: thinking about how you’re thinking

Metacognitionis the muscle of asking, mid-case: “What am I missing? Where could I be wrong? Does this story actually fit?”

Three prompts that experienced clinicians use silently:

• “What would I be embarrassed to miss here?”— the can’t-miss check.
• “What doesn’t fit my current diagnosis?” — the disconfirming-evidence check.
• “If this came back tomorrow no better, what would I do?” — the safety-net check.

Train these on every case for six months and they become automatic. They’re the unsexy version of being a good diagnostician.

How to practise clinical reasoning

Reading textbooks doesn’t build illness scripts. Cases build illness scripts. Specifically, cases where you have to commit to a diagnosis, then find out whether you were right, then think about why you were right or wrong. That feedback loop is the actual learning mechanism.

• Active commitment. Before reading the discussion in a case-based learning resource, write down your three-item differential. Reading without committing trains nothing.
• Bedside teaching. When someone presents to you, force yourself to think out loud. Articulating reasoning exposes the holes.
• Simulation. Practise on simulated patients where you can repeat the same case after feedback. Repetition with feedback is what burns illness scripts into long-term memory.
• Reflect on diagnostic surprises.Every time the diagnosis isn’t what you expected, write half a paragraph on what you missed and what would have caught it earlier. After a year, that file is the most useful textbook you own.

Pair this article with the guide to focused history-taking: a good history is the input your reasoning machinery needs.

Quick FAQ

What is a differential diagnosis?

The ranked list of conditions that could plausibly explain a patient’s presentation. Usually 3–5 likely diagnoses plus 1–2 can’t-miss possibilities that must be ruled out even if less likely.

What is dual process theory?

A model with two modes of thinking. System 1 is fast and pattern-based. System 2 is slow and analytic. Experts move between them and deliberately slow down when something feels off.

What is an illness script?

A clinician’s compressed model of a disease: who gets it, how it presents, the time course, the classic findings, what would refute it. Building good illness scripts is what experience really is.

How do I avoid premature closure?

Habitually ask “what else could this be?” — especially when the first diagnosis felt easy. Force at least three alternatives. Most diagnostic errors come from stopping too soon, not from gaps in knowledge.