What is still open

How many people actually get withdrawal?

For decades the guidelines said antidepressant withdrawal was "mild and self-limiting, lasting about a week". That collapsed: NICE revised its guidance in 2022 and the Royal College of Psychiatrists issued a new position in 2020. That much is agreed — the old reassurance was wrong.

How wrong remains open:

Both sides agree on the practical conclusion: stopping is done gradually and with a plan, never abruptly.

[12] Davies & Read 2019 · [13] Lancet Psychiatry 2024 · [14] Psychological Medicine 2025;55:e191

Tapering slowly, on its own, may not be enough

This is the newest and methodologically strongest finding in the file, and it complicates the advice to "just reduce the dose slowly". A network meta-analysis of 76 randomised trials with 17,379 participants compared five discontinuation strategies in people in remission from depression.

Note what was measured: the outcome here is not withdrawal symptoms — it is relapse of depression. That is a different question from the one above.

The confidence interval on that last row crosses 1.00: slow tapering without accompanying psychological support was not statistically different from abrupt discontinuation in terms of relapse. Slow tapering together with psychological support, by contrast, was as effective as staying on the drug.

The practical implication is an uncomfortable one for underfunded health systems: therapy is not a nice extra alongside discontinuation — it appears to be the ingredient that makes it work. And it is precisely what tends not to be covered.

[21] Zaccoletti D, Mosconi C, Gastaldon C et al. Lancet Psychiatry 2026;13(1):24–36 · doi:10.1016/S2215-0366(25)00330-X

Knowing the problem is not the same as knowing the fix

A randomised trial across 15 primary care clinics had consultant pharmacists propose tapering plans for 1,107 older adults taking benzodiazepines. Exposure fell 11.4% in the intervention clinics versus 1.5% in the controls — but the difference did not reach statistical significance (p=0.07). Falls were not reduced either.

Only in the subgroup on a low baseline dose did the intervention achieve a significant reduction (p=0.002). In other words: the deeper the dependence, the less a well-meant recommendation achieves.

We include this because it shows the scale of the problem honestly. Older, smaller educational studies reported discontinuation rates as high as 100%; the more rigorously designed modern trials produce modest to null results. Prevention is far easier than getting people off — which is why the MHRA's instruction to agree an exit plan before the first prescription matters as much as it does.

[22] Busby-Whitehead J et al. JAMA Netw Open, 26.02.2026 · doi:10.1001/jamanetworkopen.2025.60581 · [23] JAGS 2025, systematic review of deprescribing · doi:10.1111/jgs.19512

The "chemical imbalance": what was examined, and what was found

In 2022 Molecular Psychiatry published an umbrella review — a review of existing reviews and meta-analyses, not a new study in patients. It included 17 studies and examined six areas: concentrations of serotonin and its metabolite 5-HIAA in body fluids; 5-HT1A receptor binding; levels of the serotonin transporter (SERT) measured by imaging or post-mortem; tryptophan depletion studies; and associations of the SERT gene with environmental stress.

The conclusion, verbatim from the abstract: the main areas of research "provide no consistent evidence of there being an association between serotonin and depression, and no support for the hypothesis that depression is caused by lowered serotonin activity or concentrations".

But the abstract does not end there. Its next and final sentence — rarely reproduced — reads: "some evidence was consistent with the possibility that long-term antidepressant use reduces serotonin concentration". Framed as a possibility, resting on indirect data; it is not proof.

It needs two further qualifications, which almost never travel with it. First, the single human finding behind it concerns plasma serotonin — overwhelmingly platelet-derived, and it does not cross the blood–brain barrier. It says nothing about serotonin in the brain; there is, in any case, no method for measuring that in a living person. Second, the word "long-term" rests on no human duration data: the study was cross-sectional and recorded current use, with no comparison of short against prolonged treatment. The duration comes from experiments in rats.

What this does not mean. Absence of consistent evidence is not evidence of absence. The measures reviewed — 5-HIAA in body fluids, PET binding potential — are weak proxies for synaptic serotonin in the living brain. When the measuring instrument is weak, the correct conclusion is that no stable signal is detected — not that it has been shown there is none.

Two specific caveats that media coverage usually omits: acute tryptophan depletion does not consistently lower mood in healthy volunteers, but it does show clearer effects in people with a personal or family history of depression. And SERT binding potential on PET does not translate one-to-one into a "serotonin level": it may reflect terminal density, compensatory regulation, or prior drug exposure.

[15] Moncrieff J, Cooper RE, Stockmann T, Amendola S, Hengartner MP, Horowitz MA. Mol Psychiatry 2023;28(8):3243–3256 · doi:10.1038/s41380-022-01661-0 (Epub 20.07.2022) · PMID 35854107

The published rebuttal — and what it actually argues

The paper drew published objections in the same journal. The weightiest is signed by Sameer Jauhar and 34 further authors — 35 in total — under the title "A leaky umbrella has little value". Four charges: inconsistent unit of analysis; post-hoc alteration of the protocol to a self-"modified" GRADE, using the authors' own quality criteria instead of the original authors' interpretations; selective presentation of the tryptophan depletion studies; and oversimplification of molecular imaging.

The crucial clarification: these critics are not defending the "chemical imbalance" theory. They argue that the serotonin system is implicated in depression — a broader and weaker claim, not the same as "low serotonin causes depression".

The exchange continued: letters from Bartova et al. (2023), and in 2024 from Fountoulakis & Tsapakis, Arnone et al., and Smith, Carvalho & Solmi raising methodological objections — with a reply from Moncrieff et al. themselves. The dispute remains open.

What matters for the patient, and why this section exists: mechanism and efficacy are separate questions. The umbrella review did not examine whether antidepressants work — it examined whether depression is caused by low serotonin. Aspirin was used for decades before anyone knew how it worked. The collapse of the "chemical imbalance" marketing story is a real event; it is not proof that your treatment is useless.

[16] Jauhar S et al. (35 authors). A leaky umbrella has little value. Mol Psychiatry 2023;28(8):3149–3152 · doi:10.1038/s41380-023-02095-y · PMID 37322065 — the King's College London press release said "36 experts"; the published paper carries 35 · [52] letters 2023–2024 and Moncrieff reply