Health anxiety, answered

Cognitive behavioural therapy (CBT) and exposure and response prevention (ERP) have the deepest evidence base, and CBT is what the NHS offers. Pooled across trials, 66% of people respond to CBT and 48% reach remission — meaning they no longer meet the threshold for health anxiety.

"Best" needs one qualifier. Most of that evidence compares therapy against a waiting list. A Cochrane review found every form of psychotherapy beat waiting-list controls by a similar margin, the one exception being psychoeducation — being taught about the condition without any active therapy. And when CBT is compared against another active treatment rather than against nothing, the advantage shrinks considerably. The honest claim is narrower than a flat “CBT is best”: several structured therapies work meaningfully better than doing nothing, and CBT is simply the one tested most.

Acceptance and commitment therapy (ACT) and exposure therapy both have randomised trials in health anxiety showing real benefit. No trial has shown one approach clearly beating the others.

Yes. Meta-analyses find a large effect against control conditions — d = 1.01 in the largest review — and the benefit holds at 6 and 12 months. Pooled across trials, 66% of people respond and 48% reach remission.

The longest follow-up of any trial in this field is CHAMP, which tracked patients from UK medical clinics for eight years. CBT still held a significant advantage over standard care at that point, though the effect had shrunk by around 40% from its one-year size.

Effects measured against a waiting list run roughly twice those measured against an active comparison, so part of the apparent advantage is therapy beating nothing rather than CBT beating other therapies. And CBT is not risk-free — see what happens when treatment does not work, below.

Exposure therapy has its own randomised evidence, tested head-to-head against cognitive therapy rather than folded into a generic CBT package. Both beat a waiting list by large margins. Pooled across the two, 72% of patients responded after treatment and 68% a year later, with no significant difference between them.

The two work differently. Exposure means facing the feared situation or sensation — and not doing the checking, googling or reassurance-seeking that normally follows — until the anxiety falls on its own. Cognitive therapy works on the beliefs behind the fear. One trial found cognitive therapy changed how people judged moderate illness scenarios; exposure alone did not produce that shift.

In practice they are usually delivered together as CBT for health anxiety. Exposure and cognitive therapy have never been pitted against each other in a trial built to declare a winner.

Internet-delivered CBT has good trial evidence. In the key trial it was non-inferior to face-to-face therapy over the treatment period, and in an earlier trial two thirds of participants no longer met criteria for hypochondriasis afterwards. A 2024 meta-analysis found a moderate-to-large overall effect (g = 0.70), larger still against passive controls (g = 1.07).

But in every one of these trials, "internet-delivered" meant a programme with therapist guidance, feedback or supervision. No app for health anxiety has been through a clinical trial, ours included. NICE looked at digital therapies for anxiety disorders and recommended several for other conditions, but placed the health anxiety programmes in a research-only category, meaning the evidence is not yet there for routine NHS use.

The evidence supports guided digital CBT. It does not support the idea that any app you download will work, which is not something we claim.

The evidence for medication is much thinner than for therapy. The clearest trial found fluoxetine beat placebo on response rates at 12 weeks (62.5% versus 33.3%), holding to 24 weeks. A three-arm trial comparing CBT, paroxetine and placebo found response rates of 45%, 30% and 14% respectively.

Pooled across trials, SSRIs show a small but real effect against placebo — from just three trial arms and 193 participants, against 12 arms and more than 1,200 participants for CBT.

A large trial testing CBT, medication and both together found roughly half of participants did not respond to any of them. Medication decisions belong with a doctor.

Most trialled CBT protocols run roughly 5 to 12 sessions over about three months. In CHAMP the average was six sessions.

Benefits hold up well, but they fade. CHAMP found a 2.98-point advantage on the Health Anxiety Inventory at one year, maintained with no further loss between years two and five, and still significant at eight years — by which point it had shrunk to 1.83 points. A meta-analysis across trials shows the same shape: a large effect immediately after treatment (g = 0.95) falling to a smaller but real one at follow-up (g = 0.34).

A few months of treatment produces large initial gains, and a meaningful part of that holds for years. It is not a cure, and the benefit does fade.

It does not work for everyone. Most sites leave this part out.

In the review built specifically to measure unwanted outcomes, around 1 in 6 people reported at least one adverse event during CBT, up to 1 in 10 ended up measurably worse than when they started, and 10 to 19% stopped before finishing. Estimates of dropout vary — a separate meta-analysis puts it near 10%, and found people leave control conditions at the same rate, so dropping out is not evidence that the treatment is at fault.

In one analysis, patients with no personality difficulties did worse on social functioning with CBT than with standard care, while every other group did better. Treatment is not uniformly helpful.

If a course of treatment is not working, or you feel worse, that is a reason to change course rather than push on alone. Speak to your GP or the clinician overseeing your care — a different therapist, a different approach, or medication may suit you better. Do not treat an app as the answer to treatment that has failed you.

Usually with a short self-report questionnaire — the Short Health Anxiety Inventory (SHAI), the Whiteley Index or the Illness Attitude Scales — asking how much you worry about illness, check your body and seek reassurance. Condri's quiz uses the SHAI.

What a score means depends on who you are being compared with. On the SHAI-14 — the 14-item version — the cut-off that best separated high from low health anxiety was 22 in a psychiatric clinic but 29 in a general-population setting. A third cut-off, 18, was proposed for judging whether someone in treatment has reached remission. The same study grouped scores as 0–27 no or mild, 28–32 moderate, and 33–42 substantial.

A score places you relative to a comparison group. It is not a diagnosis, and only a clinician can give you one.

No, and this is the most common way health anxiety research gets over-read.

Whether someone counts as having "responded" depends entirely on the threshold a study picked. The same trial has been reported in two papers using a 25% improvement threshold in one and 30% in the other — so a person who counts as a responder in one paper is not one in the other.

Matching numbers can also hide completely different definitions. Two hypochondriasis trials both report around 54% response: in one that means a clinician rated the patient much or very much improved, in the other it means scoring one standard deviation below the baseline average on a questionnaire. The agreement is a coincidence.

Before comparing two recovery rates, check three things: the same instrument, the same threshold, and whether the figure counts everyone randomised or only those who finished.

Hypochondria is no longer a clinical diagnosis. In 2013 the DSM-5 retired hypochondriasis and split it into illness anxiety disorder — high anxiety about illness with no or only mild physical symptoms — and somatic symptom disorder, where distressing physical symptoms come with an overwhelming response to them. The symptoms in the second case are real; the diagnosis is about how much they take over.

Part of the reasoning was clinical, and part was that "hypochondriac" had become an insult, which kept people from seeking help. The WHO's ICD-11 took a different route and kept hypochondriasis, classifying it alongside obsessive-compulsive disorders because of how much it runs on intrusive thoughts and compulsive checking.

Health anxiety is the umbrella term researchers and UK clinicians use for the whole pattern, whichever formal label applies.

Searching feels like it should reduce uncertainty and reliably does the opposite.

People with health anxiety do not spot threatening information faster, but they stay on it longer once they find it. The same ambiguous symptom reads as more dangerous to them than to someone without the anxiety, and that bias is as strong below the diagnostic threshold as above it.

Then there is reinforcement. Cyberchondria — anxiety driven by health-related internet use — correlates strongly with health anxiety and behaves like a safety behaviour: relief arrives sometimes but not always, and unpredictable rewards produce the most stubborn habits. A trial where treating health anxiety produced a large drop in cyberchondria suggests the anxiety drives the searching, not only the reverse.

Estimates vary widely, and the variation is explicable rather than mysterious.

The best general-population data comes from an Australian national survey: 5.7% of people meet criteria at some point in their lives, and 3.4% at the time of being asked.

Two things make wider comparison hard. The DSM-5's 2013 split broke comparability with anything published before it, so a 2010 hypochondriasis rate and a 2020 illness anxiety disorder rate are not the same measurement. And published figures mix general-population surveys with student samples and patients already attending medical clinics — where rates run far higher, because health anxiety concentrates where people are already worried about their bodies. A review across chronic illness populations found clinically significant illness-related fear in over 20% of patients.

Not in the way the fear itself suggests. Health anxiety does not damage your body the way the illnesses you are worried about would.

One finding needs its context to mean anything, and it is not ours — it comes from researchers at the Karolinska Institutet, published in JAMA Psychiatry (Mataix-Cols et al., 2024). They followed more than 4,000 people with a formal hypochondriasis diagnosis — a clinical population, more severely affected than most people who recognise themselves in the term — against 41,290 matched controls. They found a 69% higher rate of death from any cause, and death about five years earlier on average, 70 versus 75.

The risk is not the anxiety itself. Deaths from respiratory and circulatory disease were higher while cancer deaths were not, and the researchers judged most of the excess to be from potentially preventable causes. Why is not settled: the authors point to chronic stress, health behaviours and under-recognition of the condition, without singling out one mechanism.

The same study reported a suicide risk around four times higher, but that association stopped being statistically significant once co-occurring depression and anxiety were accounted for. It should not be read as an independent risk of health anxiety.

The finding is an argument for treatment, and for staying engaged with ordinary medical care rather than avoiding it.

There is no NICE guideline written for health anxiety. The closest, CG113, covers generalised anxiety disorder and panic disorder and does not name health anxiety in its scope.

In practice it is treated within NHS Talking Therapies, where CBT is the therapy offered. NHS advice is to try self-help materials, and to see a GP if things do not improve or start affecting daily life.

NICE has assessed digital therapies for anxiety disorders. It recommends specific digital CBT programmes for PTSD, social anxiety, generalised anxiety and body dysmorphic disorder — but for health anxiety it placed them in a research-only category, meaning they should be used within approved research studies rather than as routine treatment. That reflects a gap in the evidence, not a judgement that digital therapy cannot work here.