Appearance
16.4 — How a Doctor Reaches a Diagnosis
Studies of diagnostic accuracy consistently find that the history alone yields the correct diagnosis in around 75 to 80 percent of cases. Examination adds perhaps 10 to 15 percent. Investigations add the remaining 5 to 10.
Which means the most powerful diagnostic instrument in medicine is a conversation, and the most common cause of diagnostic error is not ordering the wrong test but taking an inadequate history.
And the practical consequence for you as a patient is direct: how you describe your symptoms substantially determines whether you are diagnosed correctly.
The history
And the structure is not arbitrary — each element narrows the possibilities.
The presenting complaint, in the patient's own words.
Its characteristics. For pain, the standard framework is SOCRATES: Site, Onset, Character, Radiation, Associated symptoms, Timing, Exacerbating and relieving factors, Severity.
And each of those has discriminating power.
Onset — sudden or gradual. A headache reaching maximum intensity within seconds is a subarachnoid haemorrhage until proven otherwise (Chapter 11.10). The same headache building over an hour is not.
Character — crushing chest pain suggests cardiac; sharp and worse on breathing suggests pleural (Chapter 8.2); burning suggests neuropathic or acid.
Radiation — chest pain to the jaw or arm; loin pain to the groin; gallbladder pain to the right shoulder blade (Chapter 9.7).
Exacerbating and relieving factors — pericarditis is relieved by sitting forward (Chapter 7.2); duodenal ulcer pain by food (Chapter 9.2); arterial leg pain by hanging the leg down and venous pain by elevating it (Chapter 14.4).
Then: past medical history, drug history including over-the-counter and supplements, allergies, family history, and social history.
And the social history is the one most often rushed and most often decisive. Occupation, travel, smoking, alcohol, drugs, sexual history, pets, hobbies, and who is at home.
Asbestos exposure forty years ago. A recent flight. A new kitten. A trip to a malarial area. A job with solvents. Each of these has made a diagnosis that no test would have prompted.
Systems review — a screening set of questions across each system, catching what the patient did not think to mention.
And two symptoms carry weight out of proportion to anything else.
Unintended weight loss.Night sweats that soak the bedclothes.
Together they suggest malignancy, tuberculosis, HIV or lymphoma, and their presence changes the urgency of everything.
What patients can do
Because the history is the diagnosis, how it is given matters.
Say the most worrying thing first. Consultations are short, and the concern raised as the door opens frequently arrives too late.
Give a timeline. When it started, how it has changed.
Bring a list of medications, including anything bought without prescription and any supplements.
Say what you are worried about. "I'm worried this is cancer" is useful information, not an embarrassment — it tells the doctor what needs addressing, and unaddressed fear is a common reason people leave dissatisfied even after a correct diagnosis.
Say what you have already tried.
And bring someone if the news might be difficult, because people retain very little of what is said after a serious diagnosis.
The examination
Its role has narrowed as imaging has improved, and it retains specific value.
It confirms or refutes what the history suggested.
It finds signs the patient could not report — an enlarged liver, a heart murmur, a neurological sign.
It provides immediate information where imaging would take hours.
And some findings remain more sensitive than any test. The general impression of whether someone is seriously unwell — pallor, mottling, breathing pattern, mental state — outperforms most single measurements, and it is why experienced clinicians place weight on it.
The vital signs are the core: pulse, blood pressure, respiratory rate, temperature, oxygen saturation and level of consciousness.
And respiratory rate is the most predictive and the most frequently miscounted. A rising respiratory rate is the earliest sign of deterioration in most acute illness, and it is routinely recorded as "18" without being counted. Early warning scores exist because of this, and they work by catching the drift before anything dramatic happens.
The differential diagnosis
The list of possibilities, and the discipline is in how it is ordered.
Two competing principles.
"Common things are common." The commonest cause of a symptom is usually the cause.
And the counterweight: "what must not be missed." The differential is ordered by probability and by consequence together.
So a young person with chest pain most likely has musculoskeletal pain — and pulmonary embolism, dissection and myocarditis appear on the list not because they are likely but because missing them is catastrophic.
Occam's razor — prefer one diagnosis explaining everything.
Hickam's dictum — the deliberate counter: "a patient can have as many diseases as they damn well please." In older people with multiple conditions, Occam is frequently wrong.
Illness scripts — how expertise actually works. An experienced clinician does not reason from first principles; they recognise a pattern — this presentation fits this disease's typical script. Which is fast and accurate for typical presentations and is exactly what fails for atypical ones.
Cognitive bias
Diagnostic error contributes to a substantial proportion of adverse events, and most of it is cognitive rather than knowledge-based. The doctor knew the disease; they did not consider it.
Anchoring — fixing on an initial impression and failing to revise.
Availability bias — judging likelihood by how easily an example comes to mind. A recent memorable case distorts the estimate.
Confirmation bias — seeking evidence that fits and discounting what does not.
Premature closure — stopping once a plausible answer is found. The commonest single contributor.
Diagnostic momentum — a label attached early, carried forward by everyone downstream without re-examination. "Known anxiety" is the classic, and it is why people with anxiety are diagnosed late with physical illness.
Search satisficing — finding one abnormality and stopping. The second fracture on the X-ray is missed because the first was found.
Attribution bias — attributing symptoms to a patient's known condition, obesity, alcohol use or psychiatric history. This is a documented source of harm, and it disproportionately affects people who are already disadvantaged.
And the mitigations that work:
A diagnostic timeout — pausing to ask what else this could be.
Asking "what does not fit?" rather than what fits.
Explicitly considering the worst case.
And safety-netting — telling the patient what should make them return. "If X happens, come back" is both good care and an admission that the diagnosis is provisional, which almost all diagnoses are.
Why a test result means different things in different people
This is the most important quantitative idea in clinical medicine, and it is widely misunderstood — including by clinicians.
Sensitivity — of people with the disease, the proportion the test detects. A highly sensitive test, when negative, rules the disease out.
Specificity — of people without the disease, the proportion correctly identified as negative. A highly specific test, when positive, rules it in.
But neither answers the question the patient asks, which is: "I tested positive — do I have it?"
That is the positive predictive value, and it depends on how common the disease is in the group being tested.
The worked example, because the arithmetic is what convinces.
A test with 99 percent sensitivity and 99 percent specificity. A disease affecting 1 in 10,000. Test 1,000,000 people.
100 have the disease. The test finds 99 of them.
999,900 do not. The test wrongly flags 1 percent of them — 9,999 people.
So there are 10,098 positive results, of which 99 are correct.
A positive result means about a 1 percent chance of having the disease. Despite a test that is 99 percent accurate by both measures.
Now apply the same test where the disease affects 1 in 10. The positive predictive value rises to about 92 percent.
Same test. Same accuracy. Completely different meaning.
Which is why:
Screening asymptomatic populations generates large numbers of false positives (Chapter 16.1).
Testing without a clinical reason is not neutral — it causes anxiety, further tests, and occasionally harm from those tests.
And the pre-test probability, formed from the history, is not a soft preliminary to the hard test result. It is half the calculation.
Likelihood ratios are the formal way to combine them, and the useful rule of thumb is that a likelihood ratio above 10 or below 0.1 substantially changes the probability; between 0.5 and 2 barely changes it at all.
And two more concepts worth carrying.
The threshold approach. There is a probability below which you do not test and above which you treat without testing. Testing is only useful in the middle. A test that will not change what you do should not be done, which is the principle behind most "choosing wisely" recommendations.
The incidentaloma problem. Scan enough people and you find things. Around 3 to 7 percent of abdominal CTs show an adrenal nodule (Chapter 12.4); a substantial proportion of brain MRIs in healthy people show something. Each finding starts a cascade of further tests, each with its own risks, and the great majority are harmless.
Making the diagnosis
Pattern recognition for typical presentations — fast, and accurate when the presentation is typical.
Hypothetico-deductive reasoning for atypical ones — generate hypotheses, test them.
And the diagnosis is often provisional. A working diagnosis is acted on while remaining open to revision, and the willingness to revise is what distinguishes good diagnosis from confident diagnosis.
"I don't know yet" is a legitimate position, and it is safer than a wrong label. What matters is that it comes with a plan: what will be done, what to watch for, and when to review.
And when a diagnosis does not fit — when the treatment is not working, when the pattern is unusual — the correct move is to go back to the history rather than to order more tests. The information is usually already there.
What the next page fixes
Blood tests are the commonest investigation in medicine, and most people receive their results without understanding them. Chapter 16.5 goes through a standard blood report line by line — what each value measures, what moves it, and what the pattern means.