For a decade, doctors were told Prozac clearly helps depressed kids, but scientists traced that claim to one tiny study with numbers that don’t add up
When a child with depression needs medication, most major guidelines point doctors to the same drug: Prozac. Fluoxetine, its generic name, is among the antidepressants most often prescribed to children and teenagers, and the case for Prozac for children rests largely on two big analyses published in The Lancet and The Lancet Psychiatry. A third review, by Cochrane, looked at much the same evidence and came away far less impressed. A team of psychiatrists and statisticians has now re-run all three analyses trial by trial to find out why they disagree.
Three reviews, two different answers
In 2016, a network meta-analysis in The Lancet compared antidepressants for children and adolescents and found that only fluoxetine clearly beat placebo, with an effect size of 0.51. A 2020 analysis in The Lancet Psychiatry landed on the same number. An effect size measures how far apart two groups end up, in units of the normal spread between people. Researchers usually call 0.2 small and 0.5 moderate, so both reviews pointed to a real, moderate benefit.
Both teams also rated their confidence in that fluoxetine result as very low. The authors of the new paper argue that warnings like this tend to get lost once a number travels into summaries and guidelines.
Then in 2021, a Cochrane network meta-analysis found that fluoxetine lowered depression scores by about 2.8 points more than placebo on a rating scale that runs from 17 to 113. That works out to an effect size of roughly 0.2. The Cochrane team had decided in advance that differences under 5 points would not count as meaningful, and fluoxetine fell inside that range. Same drug, similar trials, and an estimate less than half as large.
One small trial from 2006
The new paper, by Richard Lyus, Florian Naudet, Gert van Valkenhoef and Martin Plöderl in Cochrane Evidence Synthesis and Methods, traced the gap in the evidence on Prozac for children to a single study. In 2006, researchers at one clinic in Iran randomized 40 children and adolescents to fluoxetine or nortriptyline, an older tricyclic antidepressant, with 20 in each group and no placebo. That trial reported an effect size of more than 4 in favor of fluoxetine.
For comparison, the largest effect in a 2018 analysis of 21 antidepressants in adults was below 0.5. In the 2016 Lancet dataset, the 2006 trial’s effect was more than four times the next largest one. “It’s the kind of effect size you only get if you ask people whether they prefer chocolate or faeces,” Plöderl told Nature.
A network meta-analysis does not only use direct comparisons. It also estimates how two treatments compare through a third one. The 2006 trial made nortriptyline look dramatically worse than fluoxetine, and because nortriptyline had its own trials against placebo, the math pushed fluoxetine’s advantage over placebo upward. Both Lancet teams had noticed an unexplained inconsistency in exactly this triangle of comparisons. The Cochrane review never included the trial at all, because it left out studies of tricyclic antidepressants.
The researchers also checked the trial itself with INSPECT-SR, a tool for spotting untrustworthy studies, and judged it to have serious concerns. The paper does not mention ethics approval. Children who dropped out were replaced with new ones, which randomized trials rarely do. And its numbers do not fit together: to produce the reported result, each child’s depression score before and after treatment would have had to line up almost perfectly, and the effect calculated from the final scores alone was 0.93, far below the headline figure.
What happens to Prozac for children without it
When the team re-ran both Lancet analyses without the 2006 trial, the inconsistency disappeared and fluoxetine’s effect size fell by roughly half, to a level that closely matched the Cochrane estimate of about 0.2. In other words, the three reviews stop disagreeing once one 40-child study is set aside.
Andrea Cipriani of the University of Oxford, a co-author of both Lancet analyses, told Nature he agrees the 2006 data are problematic. He said the teams discussed the trial at the time but had no accepted method for excluding a study on trustworthiness grounds, and that they flagged their very low confidence in the fluoxetine result.
Getting the trial itself examined has proved harder. The authors wrote to its researchers twice and received no reply. The journal that published it has no DOI system and is not indexed in PubMed, so the study cannot be flagged on the usual platforms. The publisher told Nature it has asked the journal’s editors to look into it.
What the re-analysis cannot show
This is a re-analysis of existing data, not a new trial. It does not show that Prozac for children is useless, and it does not test safety. In every version of the analysis, fluoxetine still did better than placebo on average. The open question is whether a small average benefit is worth the side effects, and averages say little about any single child, some of whom may respond well.
The authors could not reproduce the original analyses exactly, because the Lancet teams shared only part of their code and Cochrane had not kept its full dataset. They report that the differences were minor. Two of the authors have also published earlier critiques of fluoxetine in children, which readers may want to weigh.
The authors stop short of telling doctors to stop prescribing. Instead, “guideline committees should review their recommendations, acknowledge the evidence we have clarified and go through their process of weighing up whether use can still be justified,” Lyus told Nature. Families whose child already takes Prozac for depression should not stop it abruptly, and can raise these findings with the prescriber before deciding anything. We looked at another part of the risk and benefit picture for these drugs earlier this year, when scientists traced how serotonin-raising antidepressants can worsen tinnitus.
The study received no specific funding. Florian Naudet receives public research funding for work on research integrity. Gert van Valkenhoef is an employee of Cochrane, which the authors state was not involved in the work.
Source
Lyus R, Naudet F, van Valkenhoef G, Plöderl M. “A Re-Appraisal of Three Network Meta-Analyses to Explain the Discrepancy in Findings for the Efficacy of Fluoxetine for the Treatment of Depression in Children and Adolescents.” Cochrane Evidence Synthesis and Methods, 2026.
DOI: 10.1002/cesm.70108
Also cited
Cipriani A, Zhou X, Del Giovane C, et al. “Comparative efficacy and tolerability of antidepressants for major depressive disorder in children and adolescents: a network meta-analysis.” The Lancet, 2016.
DOI: 10.1016/S0140-6736(16)30385-3
Zhou X, Teng T, Zhang Y, et al. “Comparative efficacy and acceptability of antidepressants, psychotherapies, and their combination for acute treatment of children and adolescents with depressive disorder: a systematic review and network meta-analysis.” The Lancet Psychiatry, 2020.
DOI: 10.1016/S2215-0366(20)30137-1
Hetrick SE, McKenzie JE, Bailey AP, et al. “New generation antidepressants for depression in children and adolescents: a network meta-analysis.” Cochrane Database of Systematic Reviews, 2021.
DOI: 10.1002/14651858.CD013674.pub2
