Estimating patient-specific treatment advantages in the 'Treatment for Adolescents with Depression Study'.

The 'Treatment for Adolescents with Depression Study' (TADS, ClinicalTrials.gov, identifier: NCT00006286) was a cornerstone, randomized controlled trial evaluating the effectiveness of standard treatment options for major depression in adolescents. Whereas previous TADS analyses examined primarily effect modifications of treatment-placebo differences by various patient characteristics, less is known about the modification of inter-treatment differences, and hence, patient characteristics that might guide treatment selection. We sought to fill this gap by estimating patient-specific inter-treatment differences as a function of patients' baseline characteristics. We did so by applying the 'model-based random forest', a recently-introduced machine learning-based method for evaluating effect heterogeneity that allows for the estimation of patient-specific treatment effects as a function of arbitrary baseline characteristics. Treatment conditions were cognitive-behavioural therapy (CBT) alone, fluoxetine (FLX) alone, and the combination of CBT and fluoxetine (COMB). All inter-treatment differences (CBT vs. FLX; CBT vs. COMB; FLX vs. COMB) were evaluated across 23 potential effect modifiers extracted from previous studies. Overall, FLX was superior to CBT, while COMB was superior to both CBT and FLX. Evidence for effect heterogeneity was found for the CBT-FLX difference and the FLX-COMB difference, but not for the CBT-COMB difference. Baseline depression severity modified the CBT-FLX difference; whereas baseline depression severity, patients' treatment expectations, and childhood trauma modified the FLX-COMB difference. All modifications were quantitative rather than qualitative, however, meaning that the differences varied only in magnitude, but not direction. These findings imply that combining CBT with fluoxetine may be superior to either therapy used alone across a broad range of patients.

Informations de copyright

Copyright © 2019 Elsevier Ltd. All rights reserved.