education ·
What The Antidepressant Trials Found, & What They Left Out About Stopping
The average drug-placebo gap is small. A minority respond far more than average. And the stopping story was soft-pedalled for years. All three are true.
Before anything else: do not start, stop or change a prescription because of an article, including this one. That belongs with a provider who knows your history and can watch what happens.
Two things make going it alone a bad idea. Withdrawal is your body reacting to the absence of a drug it adapted to. Rebound is the symptom the drug was holding down coming back louder than it ever was, then settling. Rebound is the one almost nobody gets warned about, and it is why people conclude they cannot manage without a pill when what they are feeling is temporary. We wrote that up in the rebound effect, and it is worth reading before this.
This article is here because the honest version of this evidence is not the one either side usually tells.
The average gap is small
Start with the number that gets quoted least.
Researchers pooled 131 placebo-controlled trials covering 27,422 people and measured what SSRIs did to the Hamilton Depression Rating Scale, the standard clinician-rated score. The drugs beat placebo by 1.94 points (95% CI 1.37 to 2.50). Before running the analysis the authors had set 3 points as their threshold for a difference that would matter clinically, and 1.94 does not reach it (Jakobsen et al., BMC Psychiatry, 2017).
The same review found serious adverse events were more common on the drug, with an odds ratio of 1.37, which works out to about 31 per 1,000 against 22 per 1,000. And every one of the 131 trials was judged at high risk of bias.
A separate analysis, run by FDA scientists across 232 trials and 73,388 participants from 1979 to 2016, landed in the same place: a mean difference of 1.75 points (Stone et al., BMJ, 2022).
Two independent teams, very different methods, both finding an average gap of around two points on a scale that runs past fifty.
The average is hiding something, and it cuts the hopeful way
Here is where most critical write-ups stop, and where this one keeps going, because that same FDA analysis did something clever.
Instead of reporting one average, the researchers asked what the distribution of responses actually looked like. It turned out to fit three groups rather than one: people who improved a lot, people who improved moderately, and people who barely improved. Average improvements in those three groups were 16.0, 8.9 and 1.7 points.
Then they looked at who landed where. Participants taking a drug were in the large-response group 24.5% of the time, against 9.6% on placebo. They were in the minimal-response group 12.2% of the time, against 21.5% on placebo.
The authors’ conclusion is the sentence to take from this whole article: about 15% of participants had a substantial antidepressant effect beyond the placebo effect.
So both of these are true at once, and they are not in tension. The average person in these trials got about two points, which is not much. And roughly one person in seven got something real and large that placebo did not give them. A mean is a bad description of a population that comes in three kinds.
Nobody yet has a reliable way to tell in advance who that one in seven is. That is the honest state of the field, and it is also the reason “try it and watch carefully” is a defensible plan rather than a cop-out.
Severity changes the arithmetic
The benefit is not evenly spread across how ill somebody is, either.
A patient-level analysis pooled individual data and found that for people starting with a Hamilton score below 23, the effect size of medication over placebo was under 0.20, which is the standard definition of a small effect. The benefit climbed with severity and crossed the UK clinical-significance threshold at a baseline score around 25 (Fournier et al., JAMA, 2010).
That paper gets quoted constantly and its size gets quoted almost never. It pooled 6 studies and 718 patients, which is small for a claim that carries this much weight. Treat it as a direction rather than a dividing line, and note that the direction agrees with the much larger FDA dataset above.
Why the published record looked better than the real one
This part is not a matter of interpretation.
Researchers obtained the FDA reviews for 12 antidepressants covering 74 registered studies and 12,564 patients, then went looking for the matching published papers. Thirty-one percent of the studies were never published at all (Turner et al., New England Journal of Medicine, 2008).
Which studies went missing was not random. Of the studies the FDA viewed as positive, 37 were published and 1 was not. Of the studies the FDA viewed as negative or questionable, 22 were not published and 11 were published in a way the authors judged conveyed a positive outcome.
Read from the journals, 94% of trials looked positive. Read from the FDA’s own files, 51% were. Comparing the two datasets, effect sizes in the published literature were inflated by 32% overall, ranging from 11% to 69% depending on the drug.
That is the single clearest documented case of why a doctor acting in complete good faith on the published evidence could still be working from an inflated picture.
The strongest counterpoint, stated fairly
The biggest analysis in this field points the other direction, and leaving it out would be exactly the dishonesty this article is complaining about.
A network meta-analysis pulled together 522 trials covering 116,477 participants and compared 21 antidepressants. All 21 were more effective than placebo, with odds ratios from 2.13 for amitriptyline down to 1.37 for reboxetine (Cipriani et al., The Lancet, 2018).
The caveats travel with it. Of the 522 trials, 46 were rated high risk of bias and 380 moderate, with only 96 low. The authors rated the certainty of the evidence as moderate to very low. And an odds ratio for response is a different measurement from a continuous change in symptom score, so this does not cancel the two-point findings above. Both things are measuring real features of the same trials.
The chemical imbalance story
Separately from whether the drugs work, there is the question of why people were told they work.
An umbrella review gathered 17 systematic reviews, meta-analyses and large dataset analyses across every major strand of serotonin research: serotonin and its metabolites in body fluids, receptor binding, transporter levels, tryptophan depletion experiments, and the genetics (Moncrieff et al., Molecular Psychiatry, 2023).
The main areas of serotonin 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.
Moncrieff et al., Molecular Psychiatry, 2023
Be careful with what that does and does not say. It is a finding about the mechanism story, not about whether the drugs do anything. Aspirin worked for a century before anyone could explain it. A drug can have a real effect and a wrong explanation attached.
What it does undercut is the sentence millions of people were given at diagnosis, the one about correcting a chemical imbalance. That sentence shaped how people understood themselves and how long they expected to stay on treatment, and the evidence for it was never there.
Where the benefit is largest: staying on after you respond
If you want the single biggest absolute number in this entire article, it is not in the acute trials. It is in what happens after somebody gets better.
Thirty-one randomised trials followed 4,410 people who had already responded to an antidepressant and were then randomly assigned to continue or to stop. Relapse was 18% on continued treatment against 41% on placebo (Geddes et al., The Lancet, 2003).
That is a 23 percentage point difference, which means roughly four people need to continue for one relapse to be prevented over about a year. By the standards of everything above, that is a large effect.
This is worth sitting with, because it complicates the simple story in both directions. It is the best argument for not stopping the moment you feel better. It is also not an argument for staying on something forever, since most of those trials ran about twelve months and the question of year five is not answered by them.
The stopping story
For years the UK and US guidelines described antidepressant withdrawal as self-limiting, typically resolving in one to two weeks. The research caught up with that claim, and it did not hold.
The best current estimate comes from a 2024 meta-analysis of 79 studies covering 21,002 patients, which did something earlier work had not: it measured what happens when people stop a placebo, so the non-specific part could be subtracted (Henssler et al., The Lancet Psychiatry, 2024).
| Stopping an antidepressant | Stopping placebo | |
|---|---|---|
| At least one discontinuation symptom | 31% | 17% |
| Severe symptoms | 2.8% | 0.6% |
In the randomised trials specifically, the difference between the two came to about 8 percentage points. The authors’ own headline figure, after accounting for the placebo effect, is that the real incidence is approximately 15%, affecting one in six to seven people who stop.
They also found it is not uniform by drug. Desvenlafaxine, venlafaxine, imipramine and escitalopram were associated with more frequent symptoms, and imipramine, paroxetine and the venlafaxine pair with more severe ones.
A separate systematic review, drawing more on surveys of people who had actually stopped, reported higher numbers: incidence across 14 studies ranging from 27% to 86% with a weighted average of 56%, and among those with withdrawal, 46% endorsing the most severe rating offered (Davies and Read, Addictive Behaviors, 2019). Its most striking finding is about duration. Of the four studies that calculated a mean, the answers were 5 days, 10 days, 43 days, and 79 weeks.
Those two reviews disagree, and the disagreement is informative rather than embarrassing. Survey-based work reaches people motivated to answer, which pushes numbers up. Trial-based work studies people who signed up for a trial, who are often on lower doses for shorter periods, which pushes numbers down. The truth is somewhere in that range, and the honest claim is that withdrawal is common enough to plan for and variable enough that nobody can promise you a timeline.
What both agree on is that “mild and brief for everyone” was never accurate, and that a taper is a plan rather than an afterthought.
The age gradient on suicidality
This one is safety-critical, so here it is precisely.
FDA scientists pooled 372 double-blind placebo-controlled trials covering 99,231 adults, median age 42 (Stone et al., BMJ, 2009). Risk of suicidal behaviour or ideation was strongly dependent on age:
| Age | Odds ratio vs placebo |
|---|---|
| Under 25 | 1.62 (0.97 to 2.71) |
| 25 to 64 | 0.79 (0.64 to 0.98) |
| 65 and over | 0.37 (0.18 to 0.76) |
The odds fell by about 2.6% per year of age. In plain words: in adults under 25 the risk went up, and approached what is seen in children and adolescents. In midlife it looked roughly neutral on behaviour and possibly protective on ideation. In adults 65 and over, it went down.
So the two slogans are both wrong. “Antidepressants cause suicide” ignores two of the three rows. “That was debunked” ignores the first one. If the person starting treatment is under 25, close monitoring in the early weeks is not optional caution, it is what the data asks for.
What gets said, and what gets left out
| What often gets said | What is missing from it |
|---|---|
| The science is settled, they work | Mean gaps near 2 points, a severity gradient, 31% of trials unpublished |
| It is a chemical imbalance | An umbrella review found no consistent serotonin association |
| They do not work, it is all placebo | About 15% had a substantial effect beyond placebo; all 21 drugs beat placebo on response |
| If it helps, stay on it forever | Continuation has the strongest evidence, over about 12 months, not over decades |
| Discontinuation is mild and brief | 31% versus 17% on placebo, and durations ranging to many months |
| Withdrawal ruins most people who stop | The best estimate net of placebo is about one in six to seven |
| Antidepressants cause suicide | True direction under 25. Neutral in midlife, lower in those 65 and over |
Where the design shows up
The thing worth noticing is what the three-group finding implies about people.
For decades this evidence was argued as though there were one question with one answer: do antidepressants work, yes or no. The FDA data say the question was wrong. There is not one response to these drugs. There are at least three, sitting in the same trial, averaged into a number that describes none of them.
Bodies are not interchangeable. Two people with the same diagnosis, the same score and the same prescription can be having genuinely different experiences, and neither of them is lying about it. The person who got nothing is not weak and the person who got everything is not fooled.
Which is also why the terrain matters, and why it is never either-or. Sleep, food, movement, sunlight, people who know you, work that means something: none of those are a treatment for major depression and we will not pretend they are. They are the ground the treatment has to work on, and they are the part nobody else is going to fix for you.
“The LORD is near to the brokenhearted, and saves those who are crushed in spirit.” Psalm 34:18, NASB
The verse carries the conviction. It does not carry the evidence, and anyone who hands it to you instead of care is failing you.
What to take to the appointment
Not a script for refusing anything. A list for making consent real.
If a prescription is being considered: ask where this episode sits on a named scale rather than on an impression, since the benefit genuinely differs by severity. Ask what the expected benefit is against therapy, against watchful waiting, or against both together. Ask what monitoring applies for your age, and say the number out loud if you or your child is under 25.
Ask for the stop plan at the start, not at the end. What the taper would look like, over what timescale, and how you would tell withdrawal apart from the depression returning. That last distinction is the one that traps people, because the two can look alike and only one of them resolves on its own. A prescriber who has thought about it in advance is worth a great deal.
If you are already on one and doing well, that is information. You may be in the group this worked for. Nothing in this article is a reason to disturb that, and the relapse-prevention numbers above are a serious reason not to.
And if you are thinking about coming off, the whole point of this piece is that it is doable and that it is not a thing to improvise. Tapered, with a provider, with the cause addressed, knowing in advance what the first fortnight can feel like.
Read this next: what withdrawal actually is, and why it gets misread as proof you still need the drug: the rebound effect. How a side effect becomes a second prescription: the prescribing cascade. How a drug’s benefit gets quoted to you as a percentage: your statin’s benefit. And how to read a study claim at all: should you really trust the science.
Sources
- Jakobsen JC, Katakam KK, Schou A, et al. “Selective serotonin reuptake inhibitors versus placebo in patients with major depressive disorder. A systematic review with meta-analysis and Trial Sequential Analysis.” BMC Psychiatry, 2017;17(1):58. https://pubmed.ncbi.nlm.nih.gov/28178949/
- Stone MB, Yaseen ZS, Miller BJ, Richardville K, Kalaria SN, Kirsch I. “Response to acute monotherapy for major depressive disorder in randomized, placebo controlled trials submitted to the US Food and Drug Administration: individual participant data analysis.” BMJ, 2022;378:e067606. https://pubmed.ncbi.nlm.nih.gov/35918097/
- Fournier JC, DeRubeis RJ, Hollon SD, et al. “Antidepressant drug effects and depression severity: a patient-level meta-analysis.” JAMA, 2010;303(1):47-53. https://pubmed.ncbi.nlm.nih.gov/20051569/
- Turner EH, Matthews AM, Linardatos E, Tell RA, Rosenthal R. “Selective publication of antidepressant trials and its influence on apparent efficacy.” The New England Journal of Medicine, 2008;358(3):252-260. https://pubmed.ncbi.nlm.nih.gov/18199864/
- Cipriani A, Furukawa TA, Salanti G, et al. “Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis.” The Lancet, 2018;391(10128):1357-1366. https://pubmed.ncbi.nlm.nih.gov/29477251/
- Moncrieff J, Cooper RE, Stockmann T, Amendola S, Hengartner MP, Horowitz MA. “The serotonin theory of depression: a systematic umbrella review of the evidence.” Molecular Psychiatry, 2023;28(8):3243-3256. https://pubmed.ncbi.nlm.nih.gov/35854107/
- Geddes JR, Carney SM, Davies C, et al. “Relapse prevention with antidepressant drug treatment in depressive disorders: a systematic review.” The Lancet, 2003;361(9358):653-661. https://pubmed.ncbi.nlm.nih.gov/12606176/
- Henssler J, Schmidt Y, Schmidt U, Schwarzer G, Bschor T, Baethge C. “Incidence of antidepressant discontinuation symptoms: a systematic review and meta-analysis.” The Lancet Psychiatry, 2024;11(7):526-535. https://pubmed.ncbi.nlm.nih.gov/38851198/
- Davies J, Read J. “A systematic review into the incidence, severity and duration of antidepressant withdrawal effects: Are guidelines evidence-based?” Addictive Behaviors, 2019;97:111-121. https://pubmed.ncbi.nlm.nih.gov/30292574/
- Stone M, Laughren T, Jones ML, et al. “Risk of suicidality in clinical trials of antidepressants in adults: analysis of proprietary data submitted to US Food and Drug Administration.” BMJ, 2009;339:b2880. https://pubmed.ncbi.nlm.nih.gov/19671933/
Article metadata retrieved from PubMed. Photograph from Pexels. Scripture quoted from the New American Standard Bible.
Educational content from True Health Solutions. We do not diagnose, treat, cure or prevent disease, and nothing here is a substitute for care from someone who knows your history. If you are in crisis, contact your provider or call or text 988 in the US.
Questions people ask
- How well do antidepressants work compared with placebo?
- The average gap is small. A review of 131 placebo-controlled trials in 27,422 people found SSRIs reduced the Hamilton Depression Rating Scale by 1.94 points more than placebo, below the 3-point threshold the authors had set in advance for clinical significance. A separate FDA analysis of 232 trials found a 1.75-point mean difference.
- So do they help anyone?
- Yes, and the averages hide it. The same FDA analysis of 73,388 participants found responses fell into three groups, and participants on a drug were far more likely to land in the large-response group, 24.5% against 9.6% on placebo. The authors concluded roughly 15% of participants had a substantial antidepressant effect beyond the placebo effect.
- How common is antidepressant withdrawal?
- A 2024 meta-analysis of 79 studies and 21,002 patients found at least one discontinuation symptom in 31% of people stopping an antidepressant and 17% of people stopping placebo. Netting out that non-specific effect, the authors put the incidence at approximately 15%, or one in six to seven people.
- Is it true that depression is caused by low serotonin?
- An umbrella review of 17 systematic reviews and meta-analyses found no consistent evidence of an association between serotonin and depression, and no support for the idea that depression is caused by lowered serotonin. That is a finding about the mechanism story, not proof that the drugs have no clinical effect.
- Do antidepressants increase suicide risk?
- It depends strongly on age. An FDA meta-analysis of 372 trials in 99,231 adults found the odds of suicidal behaviour or ideation were 1.62 for people under 25, 0.79 for ages 25 to 64, and 0.37 for people 65 and over. Risk went up in the young, looked roughly neutral in midlife, and went down in older adults.
