A 2006 Iran trial of 40 children is said to have skewed the fluoxetine evidence base for childhood depression, with the drug's advantage vanishing once it is excluded

The researchers found that the 2016 Lancet meta-analysis and the 2020 Lancet Psychiatry meta-analysis, which concluded fluoxetine is better than placebo for depression in children and adolescents, and the 2021 Cochrane Database of Systematic Reviews meta-analysis, which found only a 'small and unimportant' advantage over placebo and that fluoxetine was probably no better than other antidepressants, disagree because of whether a single 2006 clinical trial conducted in Iran is included. The reason for the disagreement among these meta-analyses had not previously been pinned to one specific trial; this work localizes the divergence to that study and re-runs the 2016 and 2020 analyses with its data excluded, reporting that fluoxetine's outsized benefits disappeared. The evidence comes from re-analysis of existing meta-analyses rather than a new clinical trial; the text does not report the specific effect sizes, sample sizes, confidence intervals or statistical tests from the re-analysis.

The questioned 2006 trial reported an enormous advantage for fluoxetine over another antidepressant, nortriptyline, with an effect size so large as to be unheard of in biomedical science; co-author Martin Plöderl describes it as 'the kind of effect size you only get if you ask people whether they prefer chocolate or faeces.' The work treats this anomalous effect size as a central reason to judge the trial untrustworthy, and notes that no other trial of fluoxetine for paediatric depression has found an effect size close to it. This rests on the anomalous magnitude of the effect size relative to other trials, a judgement by the researchers about the plausibility of the data; the text does not list the numerical effect size.

The researchers point out that the trial's reported data imply that in the fluoxetine group every participant's depression score changed by a similar amount, and the same was true in the nortriptyline group, an orderly shift they describe as rare in clinical trials; the trial also included no description of its ethics-approval process. These specific features of the data and the missing ethics reporting are offered as additional grounds for removing the trial from the evidence base, rather than a general claim that it was low quality. This is an inspection and interpretation of the published trial report; the text provides no raw data tables or statistical re-analysis.

On this basis, Plöderl, Naudet and colleagues argue the 2006 trial should be removed from the evidence base for fluoxetine under the protocol Cochrane has introduced for assessing whether trials are 'untrustworthy', and they call on guideline committees to review their recommendations and weigh up whether use can still be justified; they stop short of calling for clinicians to stop prescribing Prozac for depression in children. Cochrane has, over the past five years, grown sufficiently worried about the influence of flawed trials to bring in a protocol for assessing untrustworthiness and potentially excluding such trials from reviews; this work applies that new protocol to the paediatric fluoxetine evidence and explicitly separates 'guideline review' from 'stop prescribing'. This is the researchers' position and recommendation rather than new clinical outcome evidence; the text also records the response of Andrea Cipriani, co-author of the 2016 and 2020 meta-analyses, who agrees the 2006 trial data are problematic and says the teams discussed it at the time but, without the current protocol, could not find a justification for excluding it, and that they noted poor methods and risk of bias and added an appendix stating 'very low' confidence in the Prozac findings.

This work addresses the evidence-synthesis and guideline-development stage: it re-runs the 2016 and 2020 meta-analyses with the 2006 trial excluded, giving guideline committees, systematic reviewers and evidence bodies such as Cochrane a basis for review, and it applies to the specific setting of drug-treatment recommendations for depression in children and adolescents. It does not supply new clinical trial data and is not a direct instruction about prescribing, since the authors explicitly do not call for clinicians to stop prescribing Prozac for depression in children. For a reader, its value lies in showing how an evidence base changes when a trial is judged untrustworthy, and how the current protocol makes such exclusion actionable.

The text does not report the specific effect sizes, confidence intervals or statistical tests from the re-analysis, so the quantitative size of the vanished advantage cannot be judged from the available text. The authors of the questioned trial did not respond to the research team, to the preprint, or to Nature's request for comment, and the publisher MedKnow says it has asked the journal's editorial team to look into the issue and expects a delay due to issues currently faced by Iran, so the outcome of that inquiry is not yet known. In addition, the 2016 and 2020 meta-analysis teams had already noted concerns about poor methods and risk of bias and stated 'very low' confidence in the Prozac findings in an appendix, leaving open the question of how far guideline developers have already taken those warnings into account.