Why Antidepressant Withdrawal Numbers Disagree So Much

The short answer#

If you read that one in six people gets withdrawal symptoms after stopping an antidepressant, and somewhere else that it is closer to one in two, both figures can be technically correct and drawn from overlapping evidence. The wide spread, from roughly 15% to more than 40%, is mostly a story about method. The choices a research team makes, above all whether it subtracts the symptoms people also report after stopping a dummy pill, move the headline number far more than any real uncertainty about how the drugs behave. So the useful skill is not memorizing a percentage. It is learning to ask how a given number was assembled before you trust it.

Key points#

• Reported discontinuation-symptom rates range from about 15% to over 40%, and the design of each study explains most of that gap.
• The biggest single lever is placebo correction: people who stop an inert pill also get headaches, dizziness, and poor sleep, so subtracting that background gives a smaller, drug-specific figure.
• Trial participants usually took the drug for only a few months, while many real patients have taken it for years, so short trials may understate what long-term users feel.
• Slow, gradual tapering is sensible and biologically well motivated, but the ideal schedule is not yet settled by strong randomized evidence.

One body of evidence, two very different headlines#

Two peer-reviewed meta-analyses published within the same year make the point cleanly. In 2024 a team led by Henssler pooled 79 studies and roughly 21,000 people in The Lancet Psychiatry. About one in three of those who stopped an antidepressant reported at least one discontinuation symptom. But so did about one in six of those who stopped placebo. Once that placebo background was subtracted, the incidence the researchers could attribute to the drug itself fell to around 15%, roughly one in six or seven people, with severe symptoms in fewer than 3%.

The same year, a meta-analysis by Zhang and colleagues in Molecular Psychiatry pooled 35 studies and reported a far higher figure: about 43% overall, and 44% within randomized trials. Neither paper is sloppy. They disagree because they asked slightly different questions of partly different studies and treated the comparison group in different ways. Understanding four of those design choices is enough to make almost any withdrawal headline legible.

Choice one: do you subtract the placebo effect#

Stopping an inert pill reliably produces headaches, dizziness, irritability, and disturbed sleep. Some of that is expectation (the nocebo effect), some is the underlying condition returning, and some is just the ordinary static of daily symptoms that everyone has. A study that reports the raw share of people with any symptom after stopping the medication scoops up all of that background noise. A study that subtracts the placebo rate tries to isolate what the drug specifically caused.

Both approaches answer real questions, but different ones. One tells you what a patient is likely to feel. The other tells you what the chemistry is responsible for. Numbers that skip placebo correction will almost always look larger, and that single decision accounts for much of the distance between 15% and 43%.

Choice two: how long did people actually take the drug#

Most randomized trials were built to test whether an antidepressant works, not to study what happens when someone comes off after years of use. As a result, the time on the drug in those pooled trials tends to be short. A 2025 appraisal in Psychological Medicine calculated that the weighted average time on treatment across the studies in the Henssler dataset was under six months, about 23 weeks, and that in most of those studies people had taken the drug for less than three months.

That is a poor match for real life. The same appraisal notes that around half of people on antidepressants in the UK have taken them for more than a year, and a comparable share in the US for more than five years. If a longer course of treatment makes withdrawal more likely or more intense, short trials will understate what long-term users go through. This is a genuine limitation of the trial-based estimate, not a bias a reviewer can subtract away.

Choice three: how symptoms were counted#

The measuring instrument moves the number before any biology enters. A structured checklist, such as the Discontinuation-Emergent Signs and Symptoms scale, prompts people about many possible symptoms and will flag more of them than a method that waits for patients to volunteer a complaint on their own. The follow-up window matters just as much. Some trials looked for symptoms only for about two weeks after stopping, which can miss slower-onset or longer-running courses. Sensitive checklists plus long windows produce higher rates than brief spontaneous reporting, independent of the drugs.

Choice four: which studies make the cut#

The sharpest divergences come from study selection. When the Psychological Medicine group reanalyzed part of the same evidence, it landed on a higher figure partly by narrowing to five studies with 601 people and setting the placebo comparison aside. Restricting a sample to the most relevant populations can be defensible, but every inclusion rule nudges the result, and a reader who cannot see the rules cannot judge the number. That is exactly why an honest summary is a range with its assumptions stated out loud, not a single crisp percentage.

Tapering: promising, not yet proven#

Away from the numbers war, there is broad agreement on two things. Stopping abruptly tends to be worse than stopping gradually, and drugs that clear the body quickly tend to cause more trouble than those that clear slowly. The leading proposal is hyperbolic tapering: cutting the dose in progressively smaller steps because the link between dose and serotonin-transporter occupancy is curved, not straight, so equal-sized cuts do not produce equal biological changes near the bottom of the range.

The mechanism is coherent and the observational signals are encouraging. A prospective Dutch cohort using tapering strips found that withdrawal during slow, daily-step tapering was limited and smaller when the dose came down more slowly, with a short-term success rate around 71%. What is thin is head-to-head randomized evidence comparing hyperbolic tapering against conventional tapering for both symptom burden and lasting discontinuation. Motivated patients following a structured program can outperform routine care for reasons that have nothing to do with the shape of the taper. The fair reading: gradual, individualized tapering is reasonable and well motivated, while the exact schedule that best balances comfort against relapse is still being worked out.

A short checklist for the next headline#

When you meet a withdrawal-incidence figure, five questions defuse most of the confusion. Was it placebo-corrected? How long had those people taken the drug? How were symptoms measured, and for how long afterward? Which studies were kept in or left out? And is the outcome any symptom at all, or only symptoms severe enough to matter clinically? The gap between 15% and 43% lives almost entirely in those answers. Underneath the fight, the durable signal is calm and worth saying plainly: discontinuation symptoms are real, most are mild, a minority are severe and genuinely disabling, and coming off more slowly generally helps.