'Probably Anxiety': The Most Expensive Phrase in Medicine

Here is the uncomfortable thing about 'probably anxiety' as a diagnosis: it is usually correct.

Anxiety is among the most common conditions in any acute setting. It genuinely produces physical symptoms — chest tightness, palpitations, breathlessness, dizziness, tingling, a sense of doom — through entirely real physiology. A clinician who attributes a young patient's palpitations to anxiety will be right far more often than wrong. And that reliability is precisely what makes the label dangerous. A diagnosis that is usually right, instantly available, effort-free to apply, and quietly flattering to the clinician's pattern-recognition is a diagnosis that will, inevitably, sometimes be applied to a pulmonary embolism. The cost of those occasions — measured in delayed diagnoses, in damaged trust, and in patients who learnt to stop reporting symptoms — is what earns the phrase its title.

Why the label sticks so easily

The mechanics deserve honesty, because they are not stupid mechanics.

The base rates do real work: in a young person with chest symptoms and a normal-looking workup, anxiety genuinely is the most likely single explanation. The presentation cooperates: anxiety and serious cardiopulmonary disease share an overlapping symptom vocabulary, and — crueller still — having a frightening physical symptom produces anxiety, so the anxious presentation of a real disease is the norm, not the exception. The label is also self-sealing in a way few diagnoses are: distress supports it, and calm doesn't refute it; a patient's insistence that 'something is really wrong' can be read as anxiety speaking, which means the diagnosis absorbs its own counter-evidence. Very few labels in medicine can do that, and every one that can should carry a hazard marking.

And the demographics are documented rather than anecdotal: the label lands hardest on young women, on patients with any psychiatric history (whose every subsequent symptom risks being routed through the existing diagnosis — 'diagnostic overshadowing', formally), and on frequent attenders, whose tenth visit is read in the light of the nine before it rather than on its own evidence. The literature on diagnostic delay in young women with serious cardiovascular and autoimmune disease keeps finding the same upstream label. None of this requires a callous clinician. It requires only the ordinary machinery — availability, priors, pattern-matching — running unexamined.

What the phrase costs

The expense ledger has three columns.

The delayed diagnosis is the obvious one: the PE, the arrhythmia, the autoimmune disease, the endocrine disorder that spent months or years filed under anxiety while it progressed. These cases populate the diagnostic-error literature and the settlement records, and their common feature is rarely a missing test — it's that the question stopped being asked.

The trust injury is less visible and arguably costlier. A patient who feels dismissed does not simply feel sad about it; they recalibrate. They report less next time, or present later, or stop presenting. The most dangerous patient in next year's waiting room is the one who learnt this year that describing symptoms earns a psychological label — so this time they wait until they can't. Dismissal compounds: it is a risk factor the system itself manufactures.

And there's a subtler cost, to the diagnosis of anxiety itself. When 'anxiety' functions as the bin for everything unexplained, it stops functioning as a real diagnosis — which it is, with its own positive features, its own severity, and its own effective approaches. Patients who genuinely need that diagnosis made well are poorly served by a system that uses it as a shrug.

The double bind — because the opposite error is also real

Intellectual honesty requires the other column of the ledger. The clinician who never says 'probably anxiety' — who investigates every anxious-seeming presentation to the end of the corridor — is not practising safer medicine. They are practising a different kind of unsafe.

Over-investigation has its own casualty list: radiation, incidental findings that beget cascades of further tests and procedures, false positives with real complications, and — least discussed — the iatrogenic reinforcement of health anxiety itself, in which each round of 'just to be sure' testing teaches the patient that their body is a crime scene requiring repeated forensic examination. For a patient whose actual condition is anxiety, an open-ended investigative odyssey is not caution. It is harm with good intentions and excellent documentation.

So the phrase can't simply be banned. The problem was never the diagnosis; it's the route by which it gets made. 'Probably anxiety' is expensive when it arrives as a default — the explanation that remains when nothing obvious turned up and the clinician stopped. It is sound when it arrives as a conclusion — reasoned towards, not fallen into.

What the diagnosis looks like when it's made properly

The difference between the default and the conclusion is checkable. Made properly, the anxiety attribution has visible structure.

The dangerous mimics were actively considered. Not 'nothing obvious showed up', but: for this presentation, the serious alternatives were named, and each was retired by something specific — a finding, an appropriately timed test, a probability honestly assessed. Exclusion did the work, not exhaustion.

There are positive features, not just an absence. Anxiety has its own evidence: the situational pattern, the prior episodes with the same signature, the response to specific reassurance, the broader picture in the patient's life. A positive diagnosis cites them. A dismissal cites only the normal tests.

The label is held provisionally and said out loud. 'I think this is most likely anxiety, and here is what would change my mind, and here is exactly what should bring you back' is a different clinical act from 'it's just anxiety'. The safety-net isn't boilerplate; it's the mechanism that lets a wrong probabilistic call get corrected instead of compounding. Probabilistic reasoning is only safe when its error-correction loop is built in.

Both things are allowed to be true. The patient with panic disorder can also have a clot — this year, this visit. Anxiety in the history is context, never armour. The discipline is treating each presentation as evidence to be weighed, not as confirmation of the file.

And the symptom is taken seriously as experienced. 'Anxiety' done well is not 'nothing is wrong'. Something is wrong — it's producing real, distressing physiology — and naming it accurately, respectfully, with a plan, is the difference between a diagnosis and a dismissal wearing one's clothes.

What this means

'Probably anxiety' earns its title not because it's usually wrong but because of how it's usually right — cheaply, instantly, self-sealingly, and most often at the expense of the patients the system already hears least. The fix was never to abandon the diagnosis, and never to test every anxious chest to exhaustion; both of those are just different distributions of harm. The fix is procedural and unglamorous: make the label pass the same checks as any other diagnosis — mimics actively retired, positive features cited, provisionality stated, return route explicit. Said that way, with the reasoning shown, it's good medicine. Said as a full stop, it's the most expensive phrase in the building — and the bill, as always, arrives later, addressed to someone else.