Data Pooling Strategies
Efficacy analyses were performed on three sets of data, obtained using the following pooling strategies:
(A) "All Studies" – data from all 8 studies were pooled. Placebo: melancholic (n = 519; 67.0%), non-melancholic (n = 256; 33.0%). Duloxetine (40–120 mg/d): melancholic (n = 759; 66.7%), non-melancholic (n = 379; 33.3%). Duloxetine was compared with placebo in one set of analyses. In a second set of analyses using data from the 6 SSRI-controlled studies, duloxetine was compared with fluoxetine and paroxetine: Placebo: melancholic (n = 348; 67.6%), non-melancholic (n = 167; 32.4%). Duloxetine (40–120 mg/d): melancholic (n = 602; 67.8%), non-melancholic (n = 286; 32.2%). SSRI: melancholic (n = 294; 68.5%), non-melancholic (n = 135; 31.5%);
(B) "Positive Studies" – data from placebo- and duloxetine-treated patients were pooled from the 6 studies (1, 4, 5, 6,7, and 8) that demonstrated a significant advantage for duloxetine over placebo on the primary efficacy measure. Placebo: melancholic (n = 415; 67.9%), non-melancholic (n = 196; 32.1%). Duloxetine (40–120 mg/d): melancholic (n = 594; 67.4%), non-melancholic (n = 287; 32.6%);
(C) "Focus Studies" – data were pooled from the 2 studies (5 and 6) that compared duloxetine 60 mg once-daily with placebo. Placebo: melancholic (n = 171; 65.8%), non-melancholic (n = 89; 34.2%). Duloxetine (60 mg/d): melancholic (n = 157; 62.8%), non-melancholic (n = 93; 37.2%).
Strategy A facilitated assessments of differential efficacy in the largest possible data set. While the inclusion of all available data has obvious advantages, the presence of failed studies could mask differential treatment effects. If a study failed to detect an overall effect it is unlikely to help detect differential subgroup effects. Therefore strategy B essentially served as a robustness check for strategy A. Pooling strategy C facilitated assessments at the recommended target dose.