Psychiatry · Medical Literature
Do antidepressants change our interpretations of facial emotions?
Abstract
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.
Clinical relevance
Does continued maintenance antidepressant treatment, compared with placebo substitution (discontinuation), sustain positive emotional processing biases as measured by a facial emotion recognition task in adults with recurrent depression who are currently well? __SNAPSHOT__ {"studyDesign":"Pre-specified mechanistic sub-study embedded within a double-blind, individually randomised, placebo-controlled trial (the ANTLER RCT)","setting":"Primary care practices across four UK sites: London, Bristol, Southampton, and York","sampleSize":"N=478 randomised in the parent trial; n=462 completing at least one facial emotion recognition task and included in the mechanistic analysis","population":"Adults aged 18–74 years with at least two prior episodes of depression, currently well (in remission), and taking one of four antidepressants (citalopram 20 mg, sertraline 100 mg, fluoxetine 20 mg, or mirtazapine 30 mg) for more than two years. More females than males participated; age distribution was skewed toward older ages.","intervention":"Maintenance antidepressant treatment (continuation of the existing antidepressant at the same dose for 52 weeks)","comparator":"Placebo substitution (discontinuation group): antidepressant tapered over approximately 2 months and replaced with an identical placebo for the remainder of the 52-week trial","primaryOutcome":"Number of faces classified as 'happy' on a computerised facial emotion recognition task in which 45 face images morphed incrementally from happy to sad (score range: 0–45); assessed at baseline, 12 weeks, and 52 weeks","followUp":"52 weeks (assessments at baseline, 12 weeks, and 52 weeks for the facial emotion task)","fundingSource":"Government (NIHR Health Technology Assessment programme, based on parent ANTLER trial funding)"}
Key findings
Of the 462 participants completing at least one facial emotion recognition task, there was no statistically significant difference between the maintenance antidepressant and placebo substitution groups at either 12 weeks (adjusted mean difference = 0.23, 95% CI −0.5 to 1.0, p = 0.5) or 52 weeks (adjusted mean difference = 0.29, 95% CI −0.5 to 1.2, p = 0.5). The effect sizes at both time points are negligible, and the confidence intervals are narrow enough to largely exclude a clinically meaningful difference in happy face classifications between treatment arms. No NNT/NNH is calculable for a continuous cognitive outcome measure of this type. In the secondary symptom-association analyses, depressive symptoms (PHQ-9) were significantly and 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), indicating that higher depressive symptom burden was robustly associated with fewer happy face identifications. Paradoxically, anxiety symptoms (GAD-7) were positively associated with happy face classifications (β = 0.11, p = 0.047), an unexpected finding whose clinical meaning requires further investigation. Notably, depressive symptoms measured by PHQ-9 were higher in the discontinuation group at 12 weeks (maintenance: 4.0 vs. discontinuation: 6.4, p < 0.001), confirming that the antidepressant vs. placebo contrast was clinically active in terms of symptom outcomes, even though no corresponding difference in emotional processing bias was detected. No adverse event data specific to the facial emotion task sub-study were reported; adverse events in the parent ANTLER trial are reported separately.
Practice implications
This study established that discontinuing antidepressants after long-term maintenance does not measurably alter facial emotion recognition bias relative to continuing treatment, across both short-term (12-week) and long-term (52-week) follow-up in a well-powered, randomised sample. This is a null finding with respect to the cognitive neuropsychological hypothesis that sustained antidepressant use maintains positive emotional processing biases. Importantly, the study did confirm that depressive symptoms are robustly associated with negative emotional processing bias (fewer happy face identifications), validating the task's construct relevance—yet antidepressant status itself did not appear to drive this bias in this long-term maintenance population. The study did NOT establish whether emotional processing bias is modified early in acute antidepressant treatment initiation (the population here was already well and on long-term therapy), nor does it address newly treated or acutely depressed patients, who may behave differently. The population was restricted to adults in primary care in remission on only four specific antidepressants at fixed doses; findings may not generalise to patients with active depression, those on other agents or higher doses, or those in secondary/tertiary care settings. The unexpected positive association between anxiety symptoms and happy face classifications warrants cautious interpretation and further mechanistic investigation. These results do not support the routine use of facial emotion recognition as a biomarker for monitoring long-term antidepressant maintenance effects. Practice should not change on the basis of this mechanistic sub-study alone; however, the findings meaningfully challenge the extrapolation of short-term laboratory evidence on emotional bias modification to clinical maintenance treatment scenarios. Only the published abstract and indexed summaries were available for review; full paper details (including pre-registration status of this specific mechanistic analysis, adherence data granularity, and full sensitivity analyses) could not be independently verified. __LIMITATIONS__ Mechanistic sub-study design: The facial emotion task was a secondary/exploratory component embedded in the ANTLER RCT, meaning it may not have been adequately powered for the specific task outcome, and the primary ANTLER trial outcome (time to depressive relapse) was distinct from the facial emotion endpoint.,Population restriction: All participants were already in remission on long-term antidepressants; findings cannot be extrapolated to acutely depressed or treatment-naive patients, who are the primary population of interest for the cognitive neuropsychological model being tested.,Antidepressant heterogeneity: Four mechanistically distinct agents (two SSRIs, one SSRI/SNRI, one NaSSA) were pooled without adequate stratification by drug class for the facial emotion outcome, potentially obscuring drug-specific effects.,Increasing medication non-adherence over time: Growing non-compliance at successive follow-up time points was explicitly noted, which could attenuate or confound treatment group differences at 52 weeks.,Surrogate/cognitive endpoint: The facial emotion recognition task is a laboratory-derived proxy measure; its relationship to clinically meaningful outcomes (relapse prevention, functional recovery) has not been established in this sample.,Missing data: Female participants were less likely to have missing task data, and missingness varied by site; a missing-at-random assumption was required, which may not hold.,External validity: All four UK sites were included, but approximately twice as many participants came from London, raising questions about geographic and sociodemographic generalisability.,Unexpected anxiety finding: The positive association between GAD-7 scores and happy face classifications runs counter to theoretical predictions and may reflect confounding, collinearity between PHQ-9 and GAD-7, or task-specific response biases that are not fully explained.,Only abstract/indexed summaries available for full methodological review, limiting assessment of pre-registration, blinding fidelity of the sub-study, and complete sensitivity analyses. __BOTTOM_LINE_CATEGORY__ potentially_practice_informing __BOTTOM_LINE__ In adults with recurrent depression who are well and on long-term maintenance antidepressants, switching to placebo did not alter facial emotion recognition bias at 12 or 52 weeks compared with continuing antidepressant treatment, despite robust cross-sectional and longitudinal associations between depressive symptoms and negative emotional processing bias. These null findings challenge the assumption that sustained antidepressant use maintains positive emotional processing biases in remitted patients and suggest that laboratory-derived cognitive neuropsychological models may not directly translate to long-term clinical maintenance contexts. No immediate practice change is warranted from this mechanistic sub-study, but findings highlight an important gap between short-term experimental evidence and long-term clinical antidepressant action that requires further dedicated mechanistic research.