Differential diagnosis: how clinicians narrow the possibilities

A differential diagnosis is the list of conditions that could explain a patient's problem. Building it, ranking it and shrinking it is the core skill of clinical medicine, and it is the skill every Stat! puzzle exercises. This guide covers how clinicians do it and the mental traps they try to avoid.

Building the list

A good differential is broad enough to include the right answer and focused enough to be useful. Clinicians use several strategies to build one.

By anatomy. For pain, ask what lives underneath it. Right upper abdominal pain suggests the liver, gallbladder, bile ducts, right kidney, the base of the right lung and the upper bowel.

By organ system. Work through the systems that can produce the symptom. Breathlessness can come from the lungs, the heart, the blood (anemia), metabolism (acid in the blood) or anxiety.

By mechanism. Mnemonics help make sure no category is forgotten. One common version is VINDICATE:

• Vascular (blocked or bleeding vessels)
• Infection and inflammation
• Neoplasm (tumors)
• Degenerative and drug-related
• Iatrogenic (caused by medical care) and idiopathic
• Congenital
• Autoimmune and allergic
• Trauma
• Endocrine and metabolic

Ranking the list

Not every item on the list deserves equal attention. Clinicians keep three questions in mind at once:

1. What is most likely? "Common things are common." A cough with fever in winter is far more often a viral infection than a rare lung disease.
2. What is most dangerous? Some diagnoses must be actively excluded because missing them is catastrophic. For chest pain, the classic "can't miss" list includes heart attack, aortic dissection, pulmonary embolism, tension pneumothorax, cardiac tamponade and a ruptured esophagus.
3. What is treatable? A treatable condition deserves to be looked for even if it is less likely.

Thinking in probabilities

Every diagnosis on the list has a pretest probability: how likely it is before a given question or test, based on the patient, the setting and what you already know. New information pushes that probability up or down. How far it moves depends on how good the evidence is.

Sensitivity and specificity

Two numbers describe how well a test separates people with a disease from people without it:

• Sensitivity is how often the test is positive in people who have the disease.
• Specificity is how often the test is negative in people who do not have it.

Two memory aids go with them:

• SnNOut: when a highly Sensitive test is Negative, it helps rule the disease Out.
• SpPIn: when a highly Specific test is Positive, it helps rule the disease In.

The D-dimer blood test is a classic example. It is very sensitive for blood clots but not specific, because many things raise it, including infection, pregnancy, cancer and recent surgery. So a normal D-dimer is useful for ruling out a clot in someone at low risk, while a raised D-dimer on its own proves very little.

Why a positive test can still be wrong

When a disease is rare in the group being tested, even a good test produces many false positives. Imagine a condition that affects 1 in 1,000 people and a test that is 99% sensitive and 95% specific. Test 1,000 people and you would expect about one true positive and about 50 false positives. Most positive results would be wrong. That is why clinicians only order some tests when the pretest probability is high enough to make a positive result meaningful.

Splitting the list: choosing the best question

The most efficient next question is the one whose answer would change your ranking the most. Asking something that every diagnosis on your list would share wastes a turn; asking something that is present in half of them and absent in the other half cuts the list in two.

Suppose your list for an older adult with breathlessness is heart failure, pneumonia and pulmonary embolism:

• Asking about fever separates pneumonia from the other two.
• Asking about ankle swelling and breathlessness when lying flat points toward heart failure.
• Asking about recent surgery, a long journey or a swollen calf raises pulmonary embolism.

Each of those questions discriminates. "Are you short of breath?" does not, because everything on the list causes it.

Cognitive traps

Even experts make predictable reasoning errors. The best known:

• Anchoring: sticking with the first diagnosis that came to mind despite new evidence.
• Premature closure: stopping the search as soon as one diagnosis seems to fit, before checking the alternatives.
• Availability: overrating a diagnosis because a recent or memorable case comes to mind.
• Confirmation bias: seeking and weighting evidence that supports what you already think.

Simple habits help: ask "what else could this be?", check that every finding is explained, and take a short diagnostic timeout before committing.

How this plays out in Stat!

Traits is a pure exercise in splitting the list. Each guess shows six traits of the disease you named, such as its organ system and whether it is acute or chronic, each marked by how well it matches the hidden disease. A well-chosen guess eliminates whole groups of diseases at once. The Traits guide shows how to pick guesses that split the field.

Syndrome asks you to build and narrow a differential from a presentation, with investigations as your questions. The Syndrome guide covers strategy, and our guide to the clinical workup explains the process behind it.