One Bad Trial, One Big Problem: How a Flawed Study Skewed Prozac Guidance for Children

Posted on 5 October, 2026 by Mehrdad Fathi

A single small clinical trial from 2006 may have quietly shaped prescribing guidelines for antidepressants in children and adolescents — for nearly two decades.

The study at the center of the controversy

A new paper published in Cochrane Evidence Synthesis and Methods has identified one clinical trial as an outlier so extreme it may have distorted the entire evidence base for using fluoxetine (Prozac) to treat depression in young people. The trial in question enrolled just 40 children in Iran and was published in 2006. Its finding: that Prozac dramatically outperformed the antidepressant nortriptyline.

The effect size reported was, by any scientific standard, implausible. “It’s the kind of effect size you only get if you ask people whether they prefer chocolate or faeces,” says Martin Plöderl, a clinical psychotherapist at Paracelsus Medical Private University in Salzburg and a co-author of the new analysis.

Beyond the implausible magnitude, the data had a further tell: in both the Prozac group and the nortriptyline group, virtually every participant’s depression score shifted by almost exactly the same amount. That kind of uniformity almost never occurs in real clinical trials.

The trial also included no description of its ethics-approval process. When Plöderl’s team tried to contact the researchers, they received no response.

How one trial moved the needle

The 2006 study found its way into two influential meta-analyses published in The Lancet (2016) and The Lancet Psychiatry (2020), both of which concluded that Prozac is superior to placebo for treating depression in children and adolescents. Those findings have since shaped national prescribing guidelines in multiple countries.

When the new research team re-ran both meta-analyses while excluding the 2006 trial, the picture changed substantially. Prozac’s outsized advantage over placebo disappeared.

The explanation for why a separate 2021 Cochrane meta-analysis had found only a “small and unimportant” benefit from Prozac — contradicting the earlier analyses — turned out to be straightforward: that review had happened to exclude the 2006 trial, because it did not include comparison studies involving nortriptyline’s drug class.

This is what researchers sometimes call a “zombie trial”: a study with data so flawed it should be considered untrustworthy, but which continues to exert influence on the evidence base long after it should have been retired.

What the guidelines currently say

National clinical guidelines generally still recommend Prozac as a first-line treatment for depression in children. In the UK, the widely used Maudsley Prescribing Guidelines describe fluoxetine as having the “strongest current evidence for efficacy” for paediatric depression, citing the 2020 meta-analysis. NICE guidelines take a more cautious line, recommending that fluoxetine be considered only in moderate-to-severe cases and only alongside psychological therapy.

David Taylor, a psychopharmacologist at King’s College London and the lead editor of the Maudsley guidelines, acknowledged the concerns. He told Nature that the next edition of the guidelines, due in 2028, “is likely to include an acknowledgement that the evidence supporting the use of antidepressants in children is, at the very least, open to different interpretations.”

Andrea Cipriani, a University of Oxford psychiatrist who co-authored both the 2016 and 2020 meta-analyses, agreed that the 2006 trial data are problematic. His teams had noted concerns about bias and poor methodology at the time, and included a caveat in an appendix to the 2016 paper that confidence in the Prozac findings was “very low.” But without a formal protocol for excluding untrustworthy trials — which the Cochrane network has since developed — they had no clear justification for removing it from the analysis.

Plöderl and his colleagues argue that this nuance was not sufficiently explicit, and that clinical guidelines did not take it adequately into account.

What this means for patients — and clinicians

The authors of the new analysis stop short of calling for clinicians to stop prescribing Prozac for children. Richard Lyus, a consultant child and adolescent psychiatrist in Cambridge and a co-author of the study, says the immediate priority is for guideline committees to review their recommendations in light of the clarified evidence.

Jon Jureidini, a psychiatry professor at the University of Adelaide who reviewed the 2016 meta-analysis, is more direct: “The main argument for using antidepressants in children is that they are better than nothing. The evidence shows otherwise.” He also notes that the drugs carry real risks, including withdrawal effects and an associated increase in suicidal thoughts in some patients.

He adds an important caution for anyone reading this: children who are currently taking Prozac should not stop without medical guidance. Any dose reduction should be gradual and supervised by a clinician.

Elia Abi-Jaoude, a psychiatrist at the Hospital for Sick Children in Toronto, described the new paper as a “slam-dunk study” and expressed hope that it would gain wide exposure.

A systemic problem in medical evidence

What makes this case notable beyond its immediate clinical stakes is what it reveals about the vulnerability of systematic reviews to individual flawed studies. Researchers have warned for years that zombie trials distort the evidence base in medicine — but clear, traceable examples are rare.

The Cochrane network has responded to these concerns by developing a protocol for assessing whether individual trials are “untrustworthy” before including them in reviews. Under that framework, the 2006 trial should now be excluded. The question is how quickly that logic propagates into the guidelines that clinicians actually use.

As Florian Naudet, a psychiatrist at the University of Rennes and a co-author of the new analysis, puts it: this is a rare chance to see exactly how a single flawed study can corrupt a front-line medical recommendation — and to start correcting it.

Source: Nature (2026). doi: https://doi.org/10.1038/d41586-026-02769-x

References:

1. Lyus, R., Naudet, F., van Valkenhoef, G. & Plöderl, M. Cochrane Evid. Synth. Methods 4, e70108 (2026).

2. Cipriani, A. et al. Lancet 388, 881–890 (2016).

3. Zhou, X. et al. Lancet Psychiatry 7, 581–601 (2020).

4. Hetrick, S. E. et al. Cochrane Database Syst. Rev. 5, CD013674 (2021).

5. Attari, A. et al. J. Res. Med. Sci. 11, 24–30 (2006).

6. Plöderl, M. et al. J. Clin. Epidemiol. 189, 112016 (2026).