Anxiety screening

Every anxiety test on the internet ends with two things: a number, and a sentence saying the number is not a diagnosis. The sentence is true. Almost nobody explains why it is true, so it reads as legal throat-clearing and the number goes in anyway.

Here is the actual reason, and it is more interesting than the disclaimer. A screening questionnaire was designed to be wrong in a particular direction. It was built to catch nearly everyone who has the thing, accepting that it will also catch a lot of people who don't, because the people running it had a second step. If you are taking it without the second step, you are holding one half of a two-part instrument and reading it as a verdict.

The GAD-7 is the instrument nearly every anxiety self-test is descended from. Spitzer and colleagues published it in 2006, and the setting matters: primary care. A doctor has eleven minutes, a waiting room, and no time to interview everyone about worry. The scale's job was to sort — to say this one, look closer — and the "look closer" was always going to happen in the same building, that week, with a person.

Seven items, each scored 0 to 3, about the last two weeks. That is the whole thing. It does not ask how long the worry has gone on, whether it is proportionate, what it is about, what you have stopped doing because of it, or whether something in your life is currently on fire. A clinician asks all of that afterwards. A web page does not, and cannot.

Plummer and colleagues pooled twelve validation samples — 5,223 people — and reported the GAD-7's performance at its usual threshold. At a cutoff of 8: sensitivity 0.83, specificity 0.84. Those are respectable numbers for a seven-item questionnaire, and they are the numbers that get quoted.

Now put a population behind them, which is the step that gets skipped.

Generalised anxiety disorder has a twelve-month prevalence of roughly 3% of adults. Take a thousand people at that rate. Thirty have GAD, and the test finds about twenty-five of them. Nine hundred and seventy do not, and the test flags about a hundred and fifty-five of them anyway. So of roughly a hundred and eighty positive screens, about twenty-five are right — around one in seven.

Change only the question and the arithmetic moves under you. If the target is any anxiety disorder rather than GAD specifically, prevalence is about 18%, and the same test at the same cutoff is right a little over half the time. Same seven questions, same score, same person — and the meaning of a positive has changed by a factor of four, because the meaning was never in the score. It was in the population the score was drawn from.

This is worth sitting with if you have ever felt indicted by a result. A high score on a self-selected internet sample — people who went looking for an anxiety test, which is its own base rate, in both directions — is a much weaker piece of evidence than it feels like while you are reading it.

One honest correction in the other direction: the same arithmetic makes a negative screen fairly informative at low prevalence. Out of those nine hundred and seventy people without GAD, more than eight hundred screen negative, and only five of the thirty cases are missed. A low score is not proof of anything, but it carries more weight than a high one. The general form of this problem — a continuous thing with a line drawn across it — is in the cutoff problem.

Two things, and they are both real.

Change in the same person. This is what the GAD-7 is mostly used for in practice: the score before treatment and the score six weeks later. Compared against itself, the instrument is stable enough to show movement, and movement is what anyone actually wants to know. A single number against a population norm tells you much less than two of your own numbers a month apart.

Naming the parts. A good screen asks about worry that will not stop, physical tension, restlessness, irritability and avoidance as separate things, which is more precision than most people bring to "I've been anxious". Knowing that the body symptoms are the loud ones, or that avoidance is doing most of the damage, changes what you do next. That is emotional granularity applied to a specific problem.

The Anxiety Check is built on that reading — fourteen questions about the last two weeks, across the worry loop, the body, and the urge to avoid — and it is worth taking as three separate readings you could describe to someone rather than one figure to be.

Anxiety is not a malfunction. It is a forecasting system, and the reason it is loud is that being wrong in one direction used to be much more expensive than being wrong in the other. Before an exam, a scan result, a hard conversation, an unpaid invoice — being anxious is the correct response and the absence of it would be the strange finding.

The line is not drawn at intensity. It is drawn at persistence, proportion and cost, and the clinical definitions say so explicitly: more days than not, for six months or more, out of proportion to the actual likelihood or impact, difficult to control, and interfering with your life.

That last clause is the one to use on yourself, and it is behavioural rather than introspective:

• The list of things you no longer do is getting longer. Calls not made, appointments postponed, invitations declined, the motorway avoided, the email left unopened for nine days. This is the single most useful indicator, and the same one the social anxiety entry lands on for the same reason: avoidance is what turns a feeling into a shrinking life.
• Decisions have stopped closing. Not indecision — the reopening of decisions already made, at 2am, indefinitely. That is intolerance of uncertainty rather than the worry itself.
• Reassurance has a half-life. You ask, you feel better, and it wears off within a day. Then you ask again, usually of someone new.
• The body has taken over the reporting. Jaw, gut, shoulders, sleep, a heart rate you now monitor. Kroenke and colleagues found anxiety disorders in about one in five primary-care patients and substantially under-detected, and a great deal of that arrives at the doctor as a physical complaint.
• It has been like this for months, without a matching reason. A bad six weeks with a cause is a bad six weeks. A bad eight months without one is a different thing.

Two errors to avoid in equal measure. Do not let a questionnaire tell you that an ordinary hard period is a disorder — most anxiety is a proportionate response to a life, and the fix is in the life. And do not let "everyone is anxious these days" talk you out of an assessment your own list of no-longer-does is quietly arguing for.

What a screen can do: give you an organised, recent, specific account of what has been happening, in the same categories a clinician will ask about, so that the appointment starts further along.

What it cannot do: diagnose, exclude a diagnosis, distinguish generalised anxiety from panic, health anxiety, social anxiety, trauma or the thyroid problem that mimics all of them, or weigh any of that against what is actually going on in your life. Those distinctions change the treatment, which is the entire reason they are made by people rather than by scoring rules.

Go and talk to someone if it has been most days for months; if you are avoiding things that matter to you; if it is costing you sleep on a regular basis; if you are drinking to take the edge off; or if a doctor has already ruled out a physical cause and the symptoms are still there. Anxiety disorders are among the most treatable problems in medicine, and untreated they tend to consolidate rather than resolve — which is an argument for going sooner rather than a reason to panic about having waited.

And if the thoughts have moved from worry to not wanting to be here, that is not the subject of this page. Talk to someone today — a doctor, an emergency service, or a crisis line in your country.