Do antidepressants change our interpretations of facial emotions?

Cognitive neuropsychological models propose that antidepressants exert their therapeutic effects by modifying negative emotional processing biases early in treatment. However, evidence from large, long-term clinical samples is limited.

We conducted a mechanistic analysis within the Antidepressants to Prevent Relapse in Depression randomized controlled trial, which compared maintenance antidepressant treatment with placebo substitution in adults with recurrent depression who were currently well (N = 478). Participants completed a computerized facial emotion recognition task at baseline, 12 weeks, and 52 weeks, in which faces morphed from happy to sad. The primary outcome was the number of faces classified as happy (0-45). Linear and longitudinal mixed-effects models were used to compare treatment groups and examine associations with depressive (PHQ-9) and anxiety (GAD-7) symptoms.

Of the 462 participants completing at least one task, there was no evidence that discontinuing antidepressants altered performance compared with maintenance at 12 weeks (adjusted mean difference = 0.23, 95% CI -0.5 to 1.0, p = 0.5) or 52 weeks (0.29, -0.5 to 1.2, p = 0.5). Depressive symptoms were negatively associated with happy face classifications both cross sectionally (β = -0.20 per PHQ-9 point, p = 0.02) and longitudinally (β = -0.09, p = 0.05). Anxiety symptoms were positively associated with happy classifications (β = 0.11, p = 0.047).

Maintenance antidepressant treatment did not sustain positive emotional processing biases as indexed by facial emotion recognition, despite robust associations between such biases and depressive symptoms. These findings challenge the generalizability of laboratory evidence on emotional bias modification to long-term clinical treatment and highlight the need for further mechanistic research on antidepressant action.