Comparative efficacy of antidepressants for acute adolescent major depressive disorder: a source-verified systematic review and network meta-analysis

Abstract

Background Adolescent-specific evidence on antidepressants is sparse and difficult to interpret because outcome data structures are heterogeneous, safety reporting is incomplete, and post-publication reanalyses have identified major reporting-integrity problems in influential trials. We aimed to estimate the comparative acute efficacy of antidepressants using source-verifiable, adolescent-specific arm-level endpoint data and to examine how historical trials with important data-quality limitations influenced the estimates. Methods We searched PubMed/MEDLINE, Cochrane CENTRAL, Web of Science, and ClinicalTrials.gov from inception to April 2026 for randomized controlled trials of antidepressants for acute treatment of adolescents aged 12–18 years, or trial-defined samples aged 12–17 or 13–18 years, with a primary DSM/ICD diagnosis of MDD. Eligibility for the primary efficacy analysis required source-verifiable, adolescent-specific arm-level endpoint means, standard deviations, and sample sizes at 12 weeks or earlier; pill placebo served as the network reference. Continuous depression outcomes were expressed as Hedges’ g standardized mean differences (SMDs), placing eligible validated scales on a common metric. Risk of bias was assessed with RoB 2, and influence analyses examined robustness to the exclusion of historical trials with major data-quality limitations. Results The antidepressant-only primary SMD network contained 4 parent trials and 11 randomized arms, including desipramine, amitriptyline, fluoxetine, levomilnacipran 40 mg, levomilnacipran 80 mg, vilazodone 15 mg, vilazodone 30 mg, and pill placebo. Desipramine showed a large effect versus pill placebo (SMD -1.80, 95% CI -2.57 to -1.03), but this estimate came from a small historical trial using completer endpoint data and a reported pooled/common SD. Amitriptyline (-0.17, -0.88 to 0.55), fluoxetine (-0.15, -0.39 to 0.08), levomilnacipran 80 mg (-0.15, -0.38 to 0.09), vilazodone 30 mg (-0.12, -0.33 to 0.09), levomilnacipran 40 mg (-0.11, -0.35 to 0.13), and vilazodone 15 mg (-0.02, -0.23 to 0.19) did not differ significantly from pill placebo. After excluding Klein 1998, and again after excluding both Klein 1998 and Kye 1996, no remaining antidepressant showed a statistically significant advantage over pill placebo. Conclusions Contemporary antidepressants in the primary network (the selective serotonin reuptake inhibitor [SSRI] fluoxetine, the serotonin-norepinephrine reuptake inhibitor [SNRI] levomilnacipran, and vilazodone, an SSRI and 5-HT1A receptor partial agonist) did not show statistically significant superiority over pill placebo, although the confidence intervals remained compatible with small benefit as well as no effect. The only statistically significant estimate was derived from a small historical desipramine trial with important data-quality limitations. Adolescent-specific arm-level evidence was too sparse and methodologically limited to support a stable comparative efficacy hierarchy among antidepressants.