Amygdala Reactivity, Antidepressant Discontinuation, and Relapse

Abstract

Importance: Antidepressant discontinuation substantially increases the risk of a depression relapse, but the neurobiological mechanisms through which this happens are not known. Amygdala reactivity to negative information is a marker of negative affective processes in depression that is reduced by antidepressant medication, but it is unknown whether amygdala reactivity is sensitive to antidepressant discontinuation or whether any change is related to the risk of relapse after antidepressant discontinuation.

Objective: To investigate whether amygdala reactivity to negative facial emotions changes with antidepressant discontinuation and is associated with subsequent relapse.

Design, setting, and participants: The Antidepressiva Absetzstudie (AIDA) study was a longitudinal, observational study in which adult patients with remitted major depressive disorder (MDD) and currently taking antidepressants underwent 2 task-based functional magnetic resonance imaging (fMRI) measurements of amygdala reactivity. Patients were randomized to discontinuing antidepressants either before or after the second fMRI measurement. Relapse was monitored over a 6-month follow-up period. Study recruitment took place from June 2015 to January 2018. Data were collected between July 1, 2015, and January 31, 2019, and statistical analyses were conducted between June 2021 and December 2023. The study took place in a university setting in Zurich, Switzerland, and Berlin, Germany. Of 123 recruited patients, 83 were included in analyses. Of 66 recruited healthy control individuals matched for age, sex, and education, 53 were included in analyses.

Exposure: Discontinuation of antidepressant medication.

Outcomes: Task-based fMRI measurement of amygdala reactivity and MDD relapse within 6 months after discontinuation.

Results: Among patients with MDD, the mean (SD) age was 35.42 (11.41) years, and 62 (75%) were women. Among control individuals, the mean (SD) age was 33.57 (10.70) years, and 37 (70%) were women. Amygdala reactivity of patients with remitted MDD and taking medication did not initially differ from that of control individuals (t125.136 = 0.33; P = .74). An increase in amygdala reactivity after antidepressant discontinuation was associated with depression relapse (3-way interaction between group [12W (waited) vs 1W2 (discontinued)], time point [MA1 (first scan) vs MA2 (second scan)], and relapse: β, 18.9; 95% CI, 0.8-37.1; P = .04). Amygdala reactivity change was associated with shorter times to relapse (hazard ratio, 1.05; 95% CI, 1.01-1.09; P = .01) and predictive of relapse (leave-one-out cross-validation balanced accuracy, 67%; 95% posterior predictive interval, 53-80; P = .02).

Conclusions and relevance: An increase in amygdala reactivity was associated with risk of relapse after antidepressant discontinuation and may represent a functional neuroimaging marker that could inform clinical decisions around antidepressant discontinuation.

Figure 1.. Study Design and Whole-Brain Functional Magnetic Resonance Imaging (fMRI) Results

A, Patients with remitted major depressive disorder were randomized to either undergo fMRI before and after antidepressant discontinuation or to undergo fMRI twice before antidepressant discontinuation. After discontinuation, all patients were followed up for 6 months. A group of control participants with no lifetime diagnosis of depression were assessed once only. The design enabled a cross-sectional comparison of the remitted depressed state (patients following a first scan [MA1] vs control individuals). In the patient sample, it allowed the effect of discontinuation to be compared to relapse (interaction of time point [MA1 vs second scan (MA2)] with group [discontinuation between the first and second scans (1W2) vs discontinuation after both scans (12W)] and relapse). B, Overall, the task significantly activated the amygdala across patients and control individuals. Shown is the z statistic map for the face-form contrast, with cluster-based correction with an activation threshold of z > 3.1 and a cluster-extent threshold of P < .001 applied at the whole-brain level.

Figure 2.. Region of Interest (ROI) Results

Panel (A) shows the 2 selected ROIs corresponding to left and right amygdala from the Harvard-Oxford atlas. Panel (B) shows the same ROI-averaged contrast values for patients for both time points and split by discontinuation group and relapse. Bars indicate means with standard errors. ^aIndicates post hoc paired-sample t test P < .01.

Figure 3.. Relapse Prediction

Depicted is the predictive accuracy of the relapse classifiers based on right and left amygdala regions of interest (ROI; ie, the modes of the posterior over the balanced accuracy inferred from the confusion matrix resulting from a leave-one-out cross-validation [LOOCV] procedure). The classifiers were based on the voxelwise increase in the face-form contrast estimates. Error bars indicate the 95% posterior predictive intervals, and the dotted line is the chance level.