# Anxiety: the practical brief *A short, honest translation of the evidence review. Not medical advice — see the last section.* --- ## The one habit that will correct more errors than anything else **An effect size without a named comparator is not a finding.** In the largest network analysis of social anxiety treatments — 101 trials, 13,164 participants — individual CBT measured **−1.19 against a waitlist** and **−0.56 against a psychological placebo**. SSRIs measured **−0.91 against a waitlist** and **−0.44 against a pill placebo**. Both lose about half their apparent magnitude when the control is made honest. More than 80% of CBT anxiety trials used waitlist controls, and only 17.4% were high quality. So: **any effect size you see quoted in this field without its comparator should be assumed to be the waitlist number, and mentally halved.** This applies to almost everything below. --- ## What actually works **Two things beat an appropriate placebo. That is the whole list.** **Individual CBT with exposure** and **SSRIs/SNRIs**, at roughly **−0.5 SMD** each. In that 101-trial network they were the only two intervention classes to clear their appropriate comparator. Everything else — MAOIs, benzodiazepines, group CBT, self-help, psychodynamic therapy — beat a waitlist and was never tested against anything better. Three things worth knowing before you start: - **Expect a 45–65% response rate.** A third to a half of people do not respond to first-line treatment. Knowing that in advance is better than discovering it and concluding you are the problem. - **Guided online CBT is equivalent to in-person** (g = 0.02). *Unguided* is substantially worse, and the gap is mostly completion: 86.7% finish face-to-face, 79.1% therapist-guided, **48.2% self-guided**. The guidance is the active ingredient, and it works by keeping you in the room. - **Medication gains are largely contingent on continuing it.** Relapse within a year is **36.4% on discontinuation versus 16.4% on continuation**. Beyond one year there is no evidence either way. Discontinuation symptoms occur in 31% versus 17% on placebo — a net incidence of roughly **1 in 6**, with severe symptoms in about 3%. **And an honest gap:** whether CBT gains hold past twelve months is largely *unstudied* rather than established. At ≥12 months the evidence is GAD k = 10, social anxiety k = 3, specific phobia k = 1, **OCD k = 0**, and non-significant for panic disorder. Relapse was reported in 6 of 69 trials. --- ## What to be sceptical of ### Anything sold on a biomarker Salivary cortisol panels, HRV-based "nervous system regulation" assessments, gut-microbiome anxiety testing. The evidence: - **Cortisol stress reactivity in anxiety: p = 0.28.** Null. The cortisol awakening response: r = .03. For calibration, about **25% of variance in laboratory cortisol responses is attributable to which country you are in** — a between-country gap larger than the entire pooled anxiety effect. - **Reduced HRV in anxiety is substantially a medication effect.** Unmedicated anxious patients did not differ from controls; starting antidepressants lowers it and stopping reverses it. The physiologists have separately stated that the measure "should not be misconstrued as a measure of vagal tone." - **Microbiome evidence in anxiety disorders totals three studies and 84 patients.** Probiotics: animals g = −0.47, humans **g = −0.12, non-significant** — same methods, same paper. There is also a deeper reason for scepticism. Three separate experiments have now **normalised an anxiety-related physiological measure without changing how people felt**: a drug that improved startle inhibition while failing clinically, 90% receptor occupancy with no anxiolysis, and breathing therapy that normalised CO₂ with no differential symptom benefit. **Moving the physiology does not reliably move the experience.** ### Apps as treatment Against an active or placebo app, the anxiety effect is **g = 0.10, non-significant**. Against nothing, 0.28. The difference is what expectancy and attention buy. (The stress review found the identical pattern: 0.09 against a placebo app, with real-world 30-day retention of 3.3%.) ### Supplements - **Ashwagandha** — four meta-analyses of overlapping trials report pooled estimates from −1.55 to **−6.87**. Seven standard deviations is larger than any psychiatric intervention ever demonstrated. I² of 93–98%, uniform positivity, and verified hepatotoxicity: 25 patients, **one liver transplant and three deaths**. - **CBD** — 54 trials, 2,477 participants, **no significant effect on anxiety**, with all-cause adverse events at NNTH 7. - **Magnesium** — the flagship review notes that no study used a validated stress measure. - **Kava** is the one botanical with a real signal (HAM-A 5.0 points), and it was withdrawn in Germany over rare liver injury. - **Lavender oil (Silexan)** has genuine trial data — but two authors work for the manufacturer, and in the pivotal trial the active comparator **paroxetine failed to beat placebo**. A trial where the established drug fails but the sponsor's product succeeds should lower your confidence. ### Propranolol for performance anxiety Near-universal among musicians and speakers. The total evidence across *all* anxiety disorders is **eight trials**; the largest social-anxiety study had **16 participants**. The systematic review concludes the evidence "is insufficient to support the routine use of propranolol in the treatment of any of the anxiety disorders." It may work. Nobody knows. ### "Some anxiety improves performance" Yerkes–Dodson is **forty mice**, with two per data point in the only condition that produced the famous curve; the easy condition was monotonic, and the paper contains no statistics and never uses the word "arousal." On meta-analysis, "facilitative anxiety" turns out to be self-confidence under another name: cognitive anxiety **r = −0.10**, self-confidence **r = +0.24**. ### Attention-bias training A closed question. The measure it was built on — the dot-probe — has an internal reliability of **zero across 36 variants and 9,600 participants**. The intervention went from d = 0.51 to non-significant in clinical samples to a registered replication finding Bayesian evidence for the null, including on the bias itself. --- ## Two findings that change what you do **Reassurance-seeking does not work, and this is measurable.** Across 14 RCTs and 3,828 patients, diagnostic testing intended to reassure produced null effects on illness worry (OR 0.87), anxiety (SMD 0.06) and symptom persistence (OR 0.99). **A negative scan does not settle the question, because the question was never really about the scan.** This is one of the most useful findings in the review and one of the least known — worth knowing before requesting the next test. **Anxiety costs more; it does not necessarily deliver less.** Across 58 studies and 8,292 participants, anxiety impaired attentional control **efficiency but not effectiveness** — the output holds, the price rises. This has two practical consequences. It means anxious people are often performing at standard while paying above-standard, which is exhausting and invisible. And it means the applied advice built on the opposite premise — that anxiety makes you worse at things — is usually wrong about the mechanism, even when it is sympathetic. *(The companion stress review reached the same conclusion from bioenergetics and from the collapse of ego depletion. Two literatures, neither citing the other, same distinction: the price of effort rises before the capacity for it falls.)* --- ## Things worth knowing about the category itself **Anxiety and depression are genetically almost the same thing.** The genetic correlation is **rg = 0.91** in the largest study to date. Sixty-three per cent of people with a current anxiety disorder currently also have depression. Quantitative structural models place GAD **closer to major depression than to panic or the phobias** — which is the opposite of how the diagnostic manuals group them. Two things follow. Don't over-invest in which side of a line you fall on. And don't assume a treatment developed for the neighbouring category cannot help — the boundaries are administratively real and biologically blurry. **"Anxiety starts in childhood" is true for the block and false for the disorders most adults have.** Phobias and separation anxiety peak at 8–13. **Panic disorder's median onset is 25–27, and GAD's is 30–35.** If you developed generalised anxiety in your thirties, you are not late — you are exactly on time. **On the "epidemic."** Age-standardised prevalence was flat from 1990 to 2010. UK *diagnosis* rates rose more than threefold — and the same metric fell 47.8% in a single month in April 2020, which tells you it measures services rather than illness. There is one clean study finding a real rise in English childhood anxiety (3.5% → 5.4%), and the modelled 25.6% COVID surge is contradicted by 134 same-participant cohorts showing no significant change. The defensible summary: **a real but modest rise in youth anxiety in some countries, inside a much larger rise in labelling.** --- ## When to involve a clinician Anxiety that is persistent and impairing warrants a proper assessment — the treatments that work are genuinely effective and most people who need them are not getting them. Globally, only **27.6% of people with a 12-month anxiety disorder receive any treatment** and **9.8% receive adequate treatment**. The most striking part of that data: only **41.3% perceive a need for care at all**, but two-thirds of those who do get some treatment. The main bottleneck is not access. It is that people do not recognise the thing as treatable. --- ## The honest bottom line Anxiety research has spent forty years failing to find a pathophysiology. The best-powered recent findings are nulls, confounds and reversals: no interoceptive-accuracy deficit, no HPA hyperactivity, no structural brain difference in GAD, no GABA deficit on spectroscopy, and serotonin running *higher*, not lower, in social anxiety. That sounds bleak, but it has a practical edge. It means the treatments that work do not depend on the mechanism being understood — CBT and SSRIs were both found without it — and it means you can safely disregard almost everything sold on a mechanistic story. **The two things with credible evidence are unglamorous, moderately effective, widely available, and not what most of the market is selling.** --- *Sources for every figure here are in the full review. Compiled 27 July 2026.*