Treatment guesses in the Treatment for Adolescents with Depression Study: Accuracy, unblinding and influence on outcomes.

Abstract

We evaluated the presence and impact of unblinding during the influential Treatment for Adolescents with Depression Study (ClinicalTrials.gov Identifier: NCT00006286). Our analysis was part of a Restoring Invisible and Abandoned Trials reanalysis. Treatment for Adolescents with Depression Study trialled fluoxetine, placebo, cognitive behaviour therapy or their combination, in treating adolescents with major depressive disorder. We analysed the accuracy of guesses of fluoxetine or placebo allocation, and their effects on change in Children's Depression Rating Scale-Revised at 12 weeks. Of 221 participants allocated to fluoxetine or placebo, 151 adolescents (68%) had their guess about pill-treatment-arm allocation recorded at week 6, and guesses were recorded for 154 independent evaluators, 159 parents and 164 pharmacotherapists. All of these groups guessed treatment allocation more accurately than would be expected by chance (60-66% accuracy; all p-values ⩽ 0.004). Guesses did not become more accurate between 6 and 12 weeks and were not predicted by adverse events, though event documentation was poor. Treatment guess had a substantial and statistically significant effect on outcome (Children's Depression Rating Scale-Revised change mean difference 9.12 [4.69; 13.55], β = 0.334, p < 0.001), but actual treatment arm did not (1.53 [-2.83; 5.89], β = 0.056, p = 0.489). Removing guess from the analysis increased the apparent effect of treatment arm, making it almost statistically significant at the conventional alpha-level of 0.05 (p = 0.06). For Treatment for Adolescents with Depression Study, treatment guesses strongly predicted outcomes and may have led to the exaggeration of drug effectiveness in the absence of actual effects. The integrity of double-blinding in trials should be routinely assessed and reported.