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

ObjectiveTo explain discrepant findings for fluoxetines efficacy in three influential network meta-analyses (NMAs) of treatments for pediatric depression, which led to conflicting clinical recommendations. DesignCritical appraisal and re-analysis of three published NMAs. Data sourcesNMAs published in two Lancet journals and in Cochrane, together with the trial datasets reported therein. Data synthesisWe compared efficacy estimates for fluoxetine versus placebo across NMAs. We identified and assessed an outlying trial included only in the Lancet NMAs using the INSPECT-SR instrument, and re-analysed the NMAs with and without this trial. ResultsThe larger effects reported in the Lancet NMAs (SMD -0.51, 95% CrI -0.99 to -0.03; and -0.51, -0.84 to -0.18) were driven by an inconsistent fluoxetine-placebo-nortriptyline loop, which the original NMA authors could not explain. We identified the cause of the inconsistency as a small outlier trial of fluoxetine versus nortriptyline that reported an implausibly large effect size (SMD > 4) favouring fluoxetine, and which was not included in the Cochrane NMA. Excluding this trial from the Lancet NMA datasets resolved the inconsistency and yielded efficacy estimates for fluoxetine that closely matched the Cochrane NMA (SMD -0.20, 95%CI -0.28 to -0.11). The outlier trial also showed multiple methodological concerns suggesting low trustworthiness. ConclusionDiscrepancies between the three NMAs were explained by the indirect influence of a single small trial with extreme and unreliable results. Removing this trial reconciled the Lancet NMAs with the Cochrane NMA, yielding a more reliable estimate of fluoxetine efficacy versus placebo. It also resolved the inconsistency. This case illustrates how inclusion of a single small problematic trial can substantially distort the clinically important results of NMAs. Our findings may alter the clinical risk/benefit assessment of fluoxetine for this indication. Other: No specific funding was involved in the study. KEY MESSAGES What is already known on this topic[bullet] Three network meta-analyses frequently inform clinical guideline recommendations for the use of fluoxetine to treat depression in children and adolescents. [bullet]However, these studies have discrepant findings for fluoxetine versus placebo, with the two earlier NMAs reporting an SMD approximately -0.5, and the other reporting a mean difference equivalent to an SMD of approximately -0.2. [bullet]Such discrepancies require investigation because network meta-analyses should be reproducible syntheses of the available evidence and different results may have different clinical implications. [bullet]Outlying or problematic trials in meta-analyses can distort results. What this study adds[bullet] We identified that the cause of the discrepant findings was the indirect influence of a single small study of fluoxetine versus nortriptyline that reported an unusually large effect size favouring fluoxetine. This study was not included in the more recent NMA, which excluded studies of tricyclic antidepressants. In addition to the extreme effect size, this study has other concerning features that call into question its credibility. [bullet]Our findings demonstrate how significantly a single small study with outlying results can impact the clinically important findings of NMAs. While it is known that meta-analyses can be affected by the inclusion of retracted studies, less is known about the impact of single unretracted studies with extreme findings and other concerning features. [bullet]Our experience demonstrates that researchers seeking to critically appraise studies with concerning features, and the evidence syntheses that include them, may face barriers when raising their concerns with study authors and journal editors. How this study might affect research, practice, or policy[bullet] Our findings may alter the clinical risk/benefit assessment of fluoxetine for this indication. [bullet]Our findings show how evidence from small studies with outlying results can influence the most clinically important findings of NMAs, and highlight the need for NMA methodology to include assessment for outliers and for the credibility of individual trial results.