Talk:Antidepressant

1. The lead emphasizes "strong support" without giving the magnitude or uncertainty of the average effect

The lead currently says that "The effectiveness of antidepressants for treating depression in adults has strong support", followed by the statement that all 21 drugs in the 2018 network meta-analysis were more effective than placebo for short-term adult MDD.^[3]

That is not necessarily false, but it risks making statistical superiority sound synonymous with a large or clearly clinically important benefit.

For example, a reanalysis of the Cipriani evidence base by Munkholm, Paludan-Müller and Boesen found that the effect was larger in published than unpublished trials and concluded that methodological limitations, high risk of bias, indirectness and publication bias substantially reduced confidence in the evidence; their pooled mean difference was only 1.97 points on the 17-item Hamilton Depression Rating Scale (0–52). They concluded that definitive conclusions about benefit were not justified from that evidence base alone.^[4]

Likewise, an FDA individual-participant-data analysis of 232 placebo-controlled trials found a mean drug-placebo difference of 1.75 HAMD points and estimated that roughly 15% of participants had a substantial antidepressant effect beyond placebo in the trial data analyzed.^[5]

I therefore think the lead would be more informative if it conveyed both sides of this evidence: antidepressants generally outperform placebo in trials, while the *average incremental effect is modest and the certainty and clinical significance are debated*.

2. Some methodological criticisms are framed as "critics" even though they are documented empirically

The article currently says "some contend that most studies on antidepressant medication are confounded by several biases" and later refers to "Critics" who maintain that trials are poorly designed or that publication bias inflates the results.

This wording may inadvertently turn an empirical methodological literature into a "pro vs. critics" dispute.

Selective publication is not merely a position held by critics. Turner et al. compared FDA reviews with the published literature for 74 antidepressant trials and found that 31% of the FDA-registered studies were unpublished; the published literature made 94% of trials appear positive compared with 51% according to the FDA analyses, with an overall inflation of effect size of about 32%.^[6]

Importantly, the 2022 update by Turner et al. found that reporting bias had diminished but persisted for newer antidepressants: among 30 trials, all 15 positive trials were transparently reported, whereas only 7 of 15 negative trials were reported transparently. Journal-based effect sizes were still somewhat larger than FDA-based effect sizes.^[7]

That seems better characterized as part of the evidence base itself, rather than primarily as a claim made by an opposing camp.

3. The article should distinguish statistical efficacy from patient-important outcomes more clearly

A 2023 systematic review and meta-analysis of 46 double-blind placebo-controlled RCTs found that antidepressants produced a small improvement in quality of life in primary MDD (SMD 0.22), with doubtful effects in some secondary-depression and maintenance settings. The authors also noted substantial risk of bias in the included studies.^[8]

Similarly, a 2023 meta-analysis of social functioning found a small effect in primary MDD, with the authors noting that the effect correlated strongly with symptom-rating outcomes.^[9]

I think these studies are relevant because the article currently spends considerably more time discussing changes in depression-rating scores than synthesizing whether benefits translate into quality of life and functioning.

4. Long-term treatment deserves a stronger evidence-quality caveat

The article discusses relapse-prevention trials and cites evidence that continuing antidepressants reduces relapse after discontinuation. That evidence is important, but the article should also acknowledge that relapse-prevention designs can be affected by withdrawal/discontinuation effects when people are rapidly switched from active medication to placebo.

For example, a time-to-event analysis of antidepressant relapse-prevention trials submitted to the FDA found that the drug-placebo separation was disproportionately large during the first few weeks after randomization and argued that some early relapse events in placebo groups might represent withdrawal reactions rather than loss of a long-term prophylactic effect.^[10]

This does not establish that maintenance benefits are illusory; rather, it is a reason to present the methodological issue alongside the relapse evidence instead of treating continuation-versus-discontinuation findings as straightforward measures of long-term prophylaxis.

A 2024 meta-analysis likewise concluded that the optimal duration of treatment remains uncertain and found no further reduction in relapse with treatment continued beyond six months in its modeling, while explicitly saying that additional benefit from longer treatment remains to be proven.^[11]

There is also a notable gap between trial duration and real-world treatment duration. A 2025 systematic analysis found a median antidepressant-use duration of about five years in U.S. data versus a median RCT duration of eight weeks among the trials it analyzed; only 6 of 52 trials randomized participants beyond 12 weeks and none exceeded 52 weeks.^[12]

5. Withdrawal deserves synthesis at the article level, not just a subsection

The article does mention discontinuation syndrome, but the evidence has developed considerably.

NICE currently advises that withdrawal can sometimes be severe, can last weeks or occasionally months, and may require tapering over a period of weeks or months; it specifically notes that paroxetine and venlafaxine are more likely to be associated with withdrawal symptoms.^[13] The Royal College of Psychiatrists similarly states that some patients need tapering over months or longer and that withdrawal can include symptoms that are difficult to distinguish from relapse.^[14]

A 2025 systematic review/reanalysis also highlights substantial uncertainty in the evidence base for withdrawal incidence and severity, particularly because many older studies were not designed to measure withdrawal systematically.^[15]

I am not proposing that the article endorse the more extreme claims sometimes made about withdrawal. In fact, the uncertainty here argues for careful wording. But the overall benefit/risk discussion should make clearer that withdrawal is an important part of the evidence base, not merely a technical adverse effect.

6. Sexual adverse effects are an example where the article could better synthesize regulatory and systematic-review evidence

The article already discusses sexual dysfunction, but the persistence question could be presented more systematically.

The EMA's pharmacovigilance committee recommended that product information for several SSRIs and SNRIs state that sexual dysfunction can in some cases continue after treatment has stopped.^[16]

At the same time, a 2023 systematic review concluded that reliable prevalence estimates for persistent post-SSRI sexual dysfunction could not yet be established and that a causal relationship could not be ascertained from the available evidence.^[17]

That combination — regulatory recognition of a possible persistent adverse effect together with uncertainty about frequency and causation — is exactly the sort of nuance I think the article should model consistently.

7. Serious-adverse-event reporting is another reason to be cautious about relying on the published trial literature alone

Hughes, Cohen and Jaggi compared industry-sponsored trial summaries with their corresponding journal articles and found substantial discrepancies in reporting serious adverse events. In their sample, 43.2% of serious adverse events recorded in trial summaries did not appear in the corresponding journal articles; most reported deaths and suicides were not reported in those articles.^[18]

Again, I do not think this means "all reported adverse effects are underestimated" or that antidepressants should be characterized as unsafe. It does mean that the article should be cautious about presenting the published RCT literature as though it were a completely neutral record of both benefits and harms.

8. The article could be more explicit about severity and population

NICE does not recommend routinely using antidepressants as the initial treatment for mild depression unless that is the person's preference, while recommending them as a first-line treatment option for moderate to severe depression.^[2]

That distinction matters because an overall statement that "antidepressants work for depression" combines populations for whom the evidence and treatment recommendations are different.

The article does mention this distinction later, but I think it belongs more prominently in the introductory synthesis.

Suggested approach

Rather than adding a large "criticism" section, I would suggest integrating the evidence into the existing structure:

• qualify the lead's "strong support" with the small average placebo-subtracted effect and the uncertainty over clinical significance;
• distinguish acute efficacy from evidence concerning maintenance/long-term treatment;
• treat publication bias and selective outcome reporting as evidence-base limitations rather than mainly as positions of "critics";
• synthesize withdrawal and sexual dysfunction alongside the overall risk-benefit discussion;
• use newer systematic reviews where available, while retaining the major evidence supporting efficacy;
• distinguish statistical significance, symptom-scale changes, remission, quality of life, functioning, relapse prevention, and individual response, since these are not interchangeable outcomes.

A possible neutral replacement for the lead sentence would be something along the lines of:

> "Randomized trials generally find that antidepressants provide a small average benefit over placebo for acute major depressive disorder in adults, although the magnitude of benefit varies with baseline severity, drug and outcome, and the evidence base is affected by reporting and methodological limitations."^[1][4][5][7]

I think that would better reflect the current literature without implying either that antidepressants are ineffective or that their benefits are unimportant. It would also make the article more internally consistent with its own existing discussion of publication bias, limitations, adverse effects and uncertainty.