Welkom bij St. Antonius Academie & Bohn Stafleu van Loghum
St. Antonius Academie heeft ervoor gezorgd dat je Mijn BSL eenvoudig en snel kunt raadplegen. Je kunt je links eenvoudig registreren. Met deze gegevens kun je thuis, of waar ook ter wereld toegang krijgen tot Mijn BSL. Heb je een vraag, neem dan contact op met René Schuwer.
Om ook buiten de locaties van St. Antonius Academie, thuis bijvoorbeeld, van Mijn BSL gebruik te kunnen maken, moet u zich eenmalig registreren. Dit kan alleen vanaf een computer op een van de locaties van St. Antonius Academie.
Eenmaal geregistreerd kunt u thuis of waar ook ter wereld onbeperkt toegang krijgen tot Mijn BSL.
Login
Als u al geregistreerd bent, hoeft u alleen maar in te loggen om onbeperkt toegang te krijgen tot Mijn BSL.
Differential Gaze Disengagement from Friendly and Disdainful Facial Expressions Predicts Reliable Increase of Depressive Symptoms Following Euthymia: Preliminary Evidence from an Eye-Tracking Study at 6-Month Follow-up
Auteurs:
Lara von Koch, Benedikt Reuter, Norbert Kathmann
Attentional biases to emotional information are assumed to play a crucial role in the onset and maintenance of depression. Moreover, recent studies show that biases may remain present in previously affected individuals during non-symptomatic stages even after acute depression has fully subsided. For example, in an investigation probing attentional disengagement from facial expressions of happiness, sadness, and disgust, never-depressed individuals showed speeded disengagement from disgusted expressions in comparison to happy faces, but this differential processing pattern was absent in currently euthymic individuals with a history of major depression.
Purpose
Building on these findings, the present follow-up study aimed to explore the predictive power of that previously described disengagement bias by assessing depressive symptoms in 63 initially euthymic individuals six months after they had participated in a gaze-contingent eye tracking task.
Methods
Each participants’ mean difference in saccade latency to initiate eye movements away from facial expressions of happiness and disgust was assessed at baseline, and tested for associations with self-reported depressive symptom gains six months later.
Results
The individual’s difference between these two emotion conditions when performing attentional disengagement (ADΔhappiness-disgust) significantly predicted the occurrence of a reliable increase in depressive symptom severity at six months follow-up. This effect remained significant when controlling for baseline symptom severity and lifetime history of depression. Conversely, dimensional change in depressive symptom severity was not predicted by the ADΔhappiness-disgust score.
Conclusions
We suggest that an individual difference score reflecting the ability to disengage attention from facial expressions of disgust versus happiness may be particularly useful in identifying individuals prone to experiencing reliable increases in depressive symptoms.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Introduction
Major depressive disorder (MDD) is a severe and globally prevalent mental health issue with a high risk of episodic recurrence and persistence of residual symptoms (Kessler et al., 2009). Even after receiving appropriate treatment, a large proportion of individuals once affected by depressive illness does not benefit sufficiently (Cuijpers et al., 2021). These empirical observations call for the exploration of strategies to improve existing treatments based on a better understanding of underlying pathomechanisms, and possibly by accounting for individual differences in disease-associated processes (Koster & Hoorelbeke, 2015).
Biased attention to emotional information is a core characteristic of clinical depression. Across various assessment methods, acutely depressed individuals generally show greater attentional allocation to negatively valenced stimuli, also called a mood-congruent bias (Peckham et al., 2010). Studies that have used eye tracking to directly measure overt attention reveal a two-fold bias, with increases of attention maintenance on negative stimuli and decreases of attention maintenance on positive stimuli in depressed individuals in comparison to healthy controls (Suslow et al., 2020). To name but a few examples, this pertains to the proportion of total dwell time in instances where opposing emotional categories are presented simultaneously, as well as longer / shorter saccade latencies in shifting gaze away from negative and towards positive facial images, and what is more, lowered inhibition of return towards recently explored areas in the visual field when emotional content is present. No evidence for abnormalities in early stages of attentional processing was found on the meta-analytic level (Suslow et al., 2020), i.e. differential processing patterns between depressed and non-depressed groups emerge consistently in the case of longer stimulus presentation (several seconds), but attentional biases in depression seem to be less pronounced within the first second of stimulus presentation, when facial featured are rapidly appraised in parafoveal vision and then receive first fixations and further scanning in the central visual field.
While the presence of biased attention in dysphoric and depressed states is evident, research to date has not sufficiently clarified if these aberrations reflect a transitory phenomenon tied to the presence of symptoms, or if they constitute a more stable cognitive disposition outlasting symptomatic episodes. Conclusions from a first meta-analysis of studies on attention biases following recovery from depressive episodes have to be weighed cautiously because of the heterogeneity of studied attentional components (e.g. initial orienting, maintained attention) as well as differing dependent measures (e.g., manual reaction times, eye tracking parameters). Nevertheless, aggregated data from the small body of research available to date supports the notion that maladaptive attention biases do in fact remain present in the remitted population, and that these biases might reflect stable vulnerability factors leaving the individual at elevated risk for depression recurrence (Shamai-Leshem et al., 2022).
The fact that biased attention has been observed in individuals remitted from depressive episodes has given rise to the idea of applying attentional bias modification (ABM) training specifically during this stage of the illness in order to prevent recurrence of depression – the rationale being that the cognitive aberration would still be manifest, but the individuals’ motivation and overall level of functioning would have increased, thus improving the ability to adhere to challenging training programs (Shamai-Leshem et al., 2022). At the same time, it has repeatedly been pointed out that reliable training effects critically depend on a thorough understanding of the altered mechanisms, and much of previous research was not optimized to meet this demand (Gober et al., 2021). For instance, many investigations have relied on manual reaction times as indirect measures of attentional shifts, possibly confounding attentional and motor processes, and poorly discriminating temporal stages of attention engagement, maintenance and disengagement (Yiend, 2010).
Addressing these common limitations, a number of recent studies have applied a gaze-contingent eye-movement task designed to test the Impaired Disengagement Hypothesis of Depression (Koster et al., 2011) across a range of study populations. The Impaired Disengagement Hypothesis states that difficulties to unhitch attentional resources from negative information account for prolonged processing of self-referent negative material and consequential impairment of mood. Existing studies have confirmed impaired disengagement from negative emotional material in acutely depressed and dysphoric individuals (Grafton et al., 2016; Sanchez et al., 2013; Southworth et al., 2017). In a more recent study, von Koch et al. (2023) have investigated if individuals fully remitted from at least one previous depressive episode exhibit impaired disengagement from negative emotional material in comparison to never-depressed individuals, probing reactions to facial expressions of happiness, sadness, and disgust. Unexpectedly, individuals with remitted depressive disorder showed abnormal processing of facial expressions of disgust in that they lacked a marked speeding in gaze disengagement from disgust compared to happiness, which, in turn, was pronounced in never-depressed participants. Importantly, previously depressed individuals did not show impaired disengagement from sad facial expressions, an attentional bias that had previously been observed in individuals exhibiting symptoms of acute depression and dysphoria. The authors took this pattern of findings to suggest that impaired disengagement from sad facial stimuli is a common correlate of depressive states representing a mood-congruency bias, while impaired disengagement from disdainful facial expressions such as disgust in contrast to facial cues of friendliness and acceptance may reflect a more stable cognitive disposition that carries over to non-symptomatic periods and possibly contributes to vulnerability for subsequent depression (see von Koch et al., 2023).
Authors of previous work have proposed that facial expressions of disgust – a core facial emotion display showing substantial overlap with the social perception of contempt when presented interpersonally (Aleman & Swart, 2008) – is highly relevant to the study of depressive reactivity (Surguladze et al., 2010, Sanchez et al., 2016). Sociomoral expressions of disgust have been shown to coincide with judgements of the target as “being bad” (Liu & Giner-Sorolla, 2024), and to convey to bystanders that the person in focus is to be kept at a distance for the sake of a group’s maintained wellbeing (Rozin et al., 1999). Relating to psychopathology, experimental work has identified heightened levels of self-disgust as a mediator between dysfunctional cognitions and symptomatology in depression sufferers (Overton et al., 2008).
Looking ahead, if difficulties disengaging attention from facial stimuli expressing disgust in contrast to facial expressions of happiness indeed reflected vulnerability in the sense put forth by cognitive models, not only should these occur in vulnerable individuals unaffected by current symptoms, but they should also prove useful in predicting the development or exacerbation of symptoms over time. Specifically, individuals with a more pronounced disadvantageous disengagement bias should be at greater odds to subsequently experience depressive symptoms.
As a pilot test of these assumptions, the present study was planned as a follow-up investigation to measure depressive symptoms six months after the initial assessment of attentional disengagement bias. Building on our previous cross-sectional findings, we hypothesized that the smaller the individuals’ difference in mean saccade latency to disengage from facial expressions displaying disgust and happiness (ADΔhappiness-disgust), the more depressive symptom severity would increase from baseline to follow-up six months later (Hypothesis 1). As the first hypothesis reflects a purely dimensional approach, a secondary analysis aimed at exploring the capability of the attentional disengagement bias measure to predict reliable worsening of depression severity on the individual level. Specifically, we hypothesized that the ADΔhappiness-disgust score would be positively associated with the odds of reporting reliably worsened symptoms of depression (Hypothesis 2). To test this assumption, reliable worsening was determined by employing an established method to ascertain statistically reliable change throughout the course of clinical symptoms (Jacobson & Truax, 1991).
Material & Methods
Design
The present study used a correlative longitudinal design where the individual difference in mean saccade latency to disengage attention from happy versus disgusted facial expressions served as the main predictor to determine (i) the association with dimensional change in depressive symptom severity and (ii) the chance of having experienced a clinically significant worsening of depressive symptom severity at 6-month follow-up. Time to disengage attention was defined as the time needed to start an eye movement leading away from an initially fixated area towards a different area on the screen in reaction to the presentation of a task-relevant probe. The level of depressive symptoms at the beginning of the observation period and the presence of previous history of major depression were controlled for in the analysis.
Sampling and Acquisition of Self-Report Data
The sample described in the present work represents a subset of participants initially recruited to take part in a larger cross-sectional study (N = 79, see von Koch et al., 2023), but comprises only individuals who provided written informed consent to be contacted for an additional assessment within the following year (N = 63). Participants were recruited initially through online advertisements and referrals from two outpatient treatment facilities. Individuals expressing interest in study participation underwent a remote screening protocol to acquire clinician-rated assessments of current and previous depressive illness. They were considered for participation only if they reported euthymic mood at baseline, defined as an over-the-phone clinician rating score below 9 on the Montgomery Asberg Depression Rating Scale (MADRS, Montgomery & Åsberg, 1979). Individuals were not considered further if they endorsed the screening question for any other current axis one disorder and indicated that related symptom distress currently influenced their everyday functioning. This was assessed with a modified version of the screening questionnaire from the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I, First & Gibbon, 2004). Indication of previous symptoms of depression did not lead to the exclusion of the participant. Following these initial screening steps, a certified clinician assessed the participant’s lifetime history of major depression by conducting module A of the SCID-I interview over the phone. During the same phone call, participants were also checked for additional exclusion criteria, that is, use of antidepressant medication (SSRI, SNRI, TCAs, MAO inhibitors, or anticonvulsants) at the time of the interview or during the three months prior, traumatic head injuries, severe neurological conditions or excessive recreational use of psychoactive substances.
During the baseline data acquisition session in the lab, after having provided written informed consent to participate in the study, each individual participant shared demographical information and completed the German version of the Beck Depression Inventory, revised (BDI-II, Hautzinger et al., 2006) as well as further questionnaires and rating scales that were not intended to test the hypotheses described here. The BDI-II is a well-validated measure of depressive symptom severity with good psychometric properties (Kühner et al., 2007), and a total score derived from 21 items ranging from 0 to 63.
Moving on to the experimental procedure, study participants then completed the eye-tracking task as described in the following section. Because the study was part of a larger project investigating the relationship between emotional biases, rumination, and mood changes in response to adverse life events, two further experimental tasks and a mood induction and regulation procedure were conducted subsequently. The complete session lasted approximately 3.5 h, and participants received compensation ranging from 30 to 40 €.
At exactly six months after the in-person assessment of gaze parameters, participants received an e-mail guiding them to an online survey where they were asked to enter follow-up data in a 10–20 min procedure. The questionnaire focused on current depressive symptom severity, and also contained an assessment of life changes and daily hassles experienced in the six months prior. Participants received € 10 for completion of this questionnaire. The study protocol used for study admission, baseline and follow-up data acquisition adhered to the Declaration of Helsinki and was approved by the ethical committee of the lead department where data acquisition was taking place.
Acquisition of Eye-Movement Data
Apparatus
Gaze parameters were recorded with an Eyelink2 headmount tracking device (SR Research Ltd., Mississauga, Canada) employing a 500 Hz sampling rate and infrared video-based recording of the left pupil only. Stimuli were presented on a 24-inch monitor with a screen resolution of 1920 × 1080 pixels using Presentation software version 18.0 (Neurobehavioral Systems, Inc., Berkeley, CA). Participants were seated at 60 cm distance from the monitor and asked to place their chin in a head rest during recording. Before each block of trials, a 9-point calibration was performed. This procedure was repeated if drift occurred that did not allow the gaze-contingent task flow to continue.
Experimental Task
Participants completed a modified version of the Engagement-Disengagement task developed by Sanchez et al. (2013). Building on previous findings, two emotional conditions were selected to index the variable of interest for the present study: disengagement from faces displaying happiness (17 trials) and disengagement from faces displaying disgust (16 trials). The experimental task entails additional conditions, namely disengagement from facial expressions of sadness (15 trials), engagement with facial expressions of happiness, disgust, and sadness, respectively (16 trials each), and finally, free viewing filler trials requiring neither engagement nor disengagement (48 trials), amounting to a total of 144 trials to complete the task. These were presented across six blocks in a pseudo-randomized sequence. All types of facial expressions were presented using photographs from the Radboud faces database (RaFD, Langner et al., 2010) and stemmed from 12 female and 12 male individuals with Caucasian appearance.
Each trial started with the presentation of a fixation cross for 500 ms, followed by the display of two horizontally aligned photographs: one of an emotional facial expression (e.g. disgust) and one of the corresponding neutral facial expression taken from the same person. Participants were instructed to freely view the screen as they saw fit, but were not given any additional instructions on how to process the images. After a fixed interval of 3000 ms viewing time, the display outwardly remained unchanged, and further stimulus and target presentation depended on participant gaze behavior: Once a fixation of 100 ms was recorded in a pre-defined interest area placed either in the location of the emotional or neutral face—depending on the trial type at hand (initial fixation needed to be on neutral face for trials probing engagement with emotion, and on emotional face for trials probing disengagement from emotion)—an arrowhead pointing upward or downward was presented at the center of the face on the opposite side of the screen. Participants had previously been instructed to press one of two horizontally aligned keys placed in front of them depending on the orientation of the arrowhead, and to do so as fast as possible. Correctly identifying the orientation of the probe required close attention to the arrowhead and thus provoked a gaze shift away from the initially fixated face in all trials requiring engagement or disengagement (no target was presented in filler trials). The start of this eye movement was later analyzed to determine the time to initiate a saccade (saccade latency).
Importantly, the single emotional category of interest was always paired with a neutral face within one screen display, and never were two discrepant emotional faces (e.g. disgust and happiness) paired at the same time. The transition following the 3000 ms free viewing period of each trial, i.e. the moment when the system started monitoring gaze behavior to “wait” for the first spontaneous fixation in the predetermined area initially to receive fixation, was not noticeable to the participant. Because an equal number of trials was included to test both disengagement from and engagement with emotional expressions, and these were presented in pseudo-randomized order, there was no way of knowing if the current trial required initial fixation on the neutral or the emotionally valenced facial expression for the task to continue. During disengagement trials, initial fixation of at least 100 ms had to happen in the area of the emotionally valenced face for the target to appear. On the other hand, during engagement trials, the initial fixation had to happen in the area of the neutral face, and the target was then presented in the area of the emotional expression. Engagement saccade latencies were not the focus of our current hypotheses. Nevertheless, these trials were included in the experimental paradigm to preserve pseudorandomization across the task and to ensure that required initial fixation on neutral and emotional faces was equally distributed across the entirety of trials during the gaze-contingent testing phase.
Because the overt task demand was always the same (indicate the orientation of arrowhead), participants were not aware that gaze shift latencies were examined until receiving full debriefing at the end of their visit. The location of areas to be initially fixated (left | right) was counterbalanced within trial demands (engagement | disengagement | free viewing) and emotional conditions. As mentioned before, a total of 48 trials only encompassed the 3000 ms free viewing period but did not continue to the probe classification task, requiring neither disengagement nor engagement. These filler trials were included to obscure gaze-contingent target presentation to participants and avoid the acquisition of a pattern of frequent gaze shifts. A typical sequence of a trial with the attentional demand of disengagement from disgust is presented in Fig. 1. Additional details concerning the experimental task procedure and specific stimuli can be found in the supplemental material to this article.
Fig. 1
Illustration of an exemplary trial in the task tondition requiring disengagement from disgust: The previously determined type of this trial (disengagement from disgusted face) requires the participant to perform attentional disengagement from the emotional expression (right side of the display) in order to orient towards the neutral expression (left side). Facial photographs and screen dimensions not to scale. Permission to reproduce facial images as stimulus examples was obtained by Radboud Faces Database (Langner et al., 2010)
×
Data Analysis
Clinical Outcome Variables
For the primary outcome variable indicating dimensional change of depressive symptoms, the numeric difference between baseline and follow-up BDI-II scores was calculated (gain score: baseline score was subtracted from follow-up score). For the secondary outcome variable–reliable worsening of depression symptom severity–we coded if the BDI-II score at follow-up had risen by at least 9 points compared to baseline. This difference reflects Jacobson and Truax’s reliable change criterion to determine clinically significant change (Jacobson & Truax, 1991) when entering the standard deviation from a normative sample of remitted depressed individuals (SD = 9.6) and internal consistency of α = 0.89, as published in the German version of the BDI-II testing manual (Hautzinger et al., 2006), resulting in an RC criterion of 8.83.1 The criterion determined in the present study matches previous studies investigating reliable worsening of symptoms as indexed by a score increase ≥ 9 on the BDI-II (e.g. Busch et al., 2013).
Eye-Movement Data Analysis
Analysis was performed using EyeLink DataViewer analysis software version 4.1.63 (SR Research, Mississauga, Canada). We analyzed all trials with saccades spanning at least 9° of visual angle and ending in a pre-defined interest area corresponding to the facial display area where the target had been presented. Saccade detection was based on a velocity threshold of 30° per second. Trials entered further processing if (i) there was a time lag of at least 80 ms between probe presentation and the start of the eye-movement, indicating that the gaze shift did in fact occur as a reaction to the probe (Wenban-Smith & Findlay, 1991), and (ii) the saccade latency observed on the individual trial level lay within three standard deviations from the participant’s mean across all trials. The overall proportion of included trials following these criteria was 90%. We proceeded to determine the individual’s mean saccade latency to disengage from facial expressions of disgust and happiness, respectively, that is, the average time period between the onset of the visual probe and the start of the saccade connecting the area of the face initially fixated to the area of the face where the probe was located. Internal consistency estimations for saccade latencies observed in the current sample are indicated by the following Cronbachs coefficients, and spoke to good reliability in both experimental conditions (disengagement from happiness: r = 0.90, disengagement from disgust: r = 0.89).
The ADΔhappiness-disgust score was calculated by subtracting the individual’s mean saccade latency to disengage from disgust from the individual’s mean saccade latency to disengage from happiness, so that larger scores on the difference measure speak to relatively longer processing of positively valenced images (happiness), while small or negative scores on the difference measure indicate longer visual processing of the negative facial expression (disgust).
Statistical Analysis
All statistical analyses were conducted using R version 4.0.3 (R Development Core Team, 2017), using a two-tailed α = 0.05. Receiver operating characteristic (ROC) curve analysis and visualization was produced using the library programs ROCit, version 2.2.1, (Khan & Brandenburger, 2020), and cutpointr, version 1.1.2, (Thiele & Hirschfeld, 2021).
As a test of the primary hypothesis, a generalized linear regression model was fit to predict numeric change in depressive symptom severity from the individual’s ADΔhappiness-disgust score. To test our second hypothesis, we constructed a logistic regression model to predict the binary criterion of reliable worsening of depressive symptom severity (reliable worsening present | absent) from the individual’s attentional disengagement ADΔhappiness-disgust score. In both regression models, baseline depressive symptom levels (continuous) and lifetime history of depression (present | absent) were entered as additional predictors to control for.
In addition, a ROC curve analysis was performed to determine the utility of ADΔhappiness-disgust as a classifier to predict the binary criterion of reliable worsening of depressive symptom severity (reliable worsening present | not present). Youden’s J was calculated and used as an indicator to determine a cut point on the predictor variable to optimally classify cases in this sample.
The group of participants who had experienced reliable worsening of depressive symptoms after six months were tested for differences in observed variables compared to the remaining group of participants using independent sample t-tests for continuous and χ2-tests for discrete variables. Finally, additional generalized linear and logistic regression models were constructed to conduct a posteriori analyses.
Transparency and Openness
Preprocessed data and analysis code to reproduce the reported results have been made publicly available here: https://osf.io/sxtph. Our hypotheses and analytic strategy for this manuscript were not pre-registered. We acknowledge that our sample size for the longitudinal analysis was limited by the sample size collected for a previous cross-sectional study. Given the lack of an a-priori power calculation for this step, this predictive analysis should be regarded as exploratory.
Results
Participants’ Characteristics
Participants mean age was 28.0 years (SD = 6.7), and 89% were female. The mean BDI-II score at baseline was M = 3.5 (SD = 4.3) and M = 5.7 (SD = 6.6) at follow-up. 56% of participants met criteria for lifetime depression as assessed in the SCID-I interview conducted during study inclusion. 74% of the individuals with a lifetime diagnosis reported to have experienced one episode of MDD, 17% reported two or three episodes, 9% four or more. For participants with a history of major depression, time elapsed since the end of the last depressive epsode was 3.5 years on average (SD = 2.7).
Mean overall saccade latency to disengage attention was M = 182.06 ms (SD = 43.21) for faces expressing happiness, and M = 178.18 ms (SD = 42.11) for faces expressing disgust. The mean difference between disengagement from happiness and disengagement from disgust on the individual’s level was M = 3.88 ms (SD = 22.25), indicating faster disengagement from disgust than from happiness overall.
Association Between Attentional Disengagement and Dimensional Change in Depressive Symptom Severity
Simple linear regression was used to test if ADΔhappiness-disgust significantly predicted changes in depressive symptom severity from baseline to follow up. Contrary to our first hypothesis, the overall regression with ADΔhappiness-disgust scores entered as the predictor of depressive symptom changes from baseline to follow up yielded neither a significant overall model fit (F(1,61) = 0.14, p = 0.70, R2 = 0.01), nor a statistically significant prediction by the variable ADΔhappiness-disgust (β = − 0.01, p = 0.70). Entering the clinical variables (i) depressive symptom severity at baseline and (ii) previous history of major depression (absent | present) into a multiple linear regression model as simultaneous predictors together with ADΔhappiness-disgust resulted in a significant overall model fit (F(3,59) = 8.50, p < 0.001, R2 = 0.30), but the predictive influence of ADΔhappiness-disgust remained insignificant (β = − 0.01, p = 0.71), while it was found that initial depressive symptom severity significantly predicted changes in depressive symptoms from baseline to follow-up (β = − 3.56, p < 0.001), as did a pre-existing history of major depression (β = 6.50, p < 0.001).
Association Between Attentional Disengagement and Reliable Worsening of Depressive Symptoms
Logistic regression was used to test if ADΔhappiness-disgust significantly predicted the occurrence of reliable worsening of depressive symptoms at 6-month follow-up. At first, ADΔhappiness-disgust was entered as the single predictor (model 1). In a second step, a logistic regression model predicting the occurrence of a reliable increase in symptoms from ADΔhappiness-disgust, depressive symptom severity at baseline and previous history of major depression (model 2) was fitted. Results are displayed in Table 1.
Table 1
Results of the logistic regression analyses predicting the occurrence of reliable worsening of depressive symptoms at 6-month follow-up from ADΔhappiness-disgust gaze parameter and clinical measures (N = 63)
Prediction of reliable worsening
VIF
Baseline measures entered
R2
χ2
p
OR
95% CI
Model 1
.21a
6.85
.009
ADΔhappiness-disgust
.023
0.94
[0.89—0.99]
-
Model 2b
.49a
10.95
.004
ADΔhappiness-disgust
.016
0.93
[0.87 – 0.99]
1.09
lifetime MDDc
.006
–d
–d
1.00
BDI-II at baseline
.228
0.62
[0.27 – 1.38]
1.09
Logistic regression was used to predict the dichotomous variable (presence of reliable increase in depressive symptoms) at 6-month follow-up (0 = absent, 1 = present). OR = odds ratio; 95% CI = 95% confidence interval, VIF = variance inflation factor. ADΔhappiness-disgust = difference score in attentional disengagement from happy vs. disgusted facial expressions; BDI-II = Beck Depression Inventory, revised
aNagelkerke R2 and χ2 statistics are reported
bFirth method using penalized maximum likelihood was used to parametrize this model. This method was chosen to account for the fact that no cases with reports of reliable symptom increase were observed amongst individuals without previous history of depression, resulting in a quasi-complete separation of cases by this single predictor variable
cPresent vs. absent
dBecause of quasi-complete separation (see b), no interpretable odds ratio can be reported for this predictor
Cases with reliable worsening of depressive symptoms were observed in 11% of participants, and all of the affected individuals happened to have reported a history of MDD prior to study inclusion. A summary of subgroup differences between individuals affected by a reliable symptom increase (n = 7) and those unaffected (n = 56) can be found in Table 2.
Table 2
Demographic, clinical, and eye tracking data for the subgroups of participants affected/unaffected by reliable worsening of depressive symptoms at 6-month follow-up
Study variables
Reliable worsening a
n = 7
No reliable worsening b
n = 56
Group comparison statistic
Age
27.57 (3.46)
27.85 (6.73)
t(13) = 0.18,
p = .862
Gender ratio (n female | n male)
5 | 2
49 | 7
Χ2(1) = 0.33,
p = .567
Education ratio (n high-school diploma | n university entrance diploma)
2 | 5
3 | 53
Χ2(1) = 1.96,
p = .161
BDI-II at baseline
3.4 (4.1)
3.5 (4.4)
t(8) = 0.02,
p = .983
BDI-II at follow-up
19.1 (8.4)
4.0 (4.4)
|t|(6) = 4.69,
p = .002
Number of MDEs prior to study inclusion
1.71 (1.11)
0.73 (1.04)
Χ2(5) = 13.48,
p = .019
Saccade latency to disengage from happiness
162.89 (45.83)
184.45 (42.70)
t(7) = 1.18,
p = .274
Saccade latency to disengage from disgust
176.95 (41.66)
178.33 (49.11)
t(7) = 0.07,
p = .945
ADΔhappiness-disgust
-14.06 (8.51)
6.12 (22.45)
t(19) = 4.60,
p < .001
Values displayed indicate arithmetic means and standard deviations in parentheses, unless stated otherwise. BDI-II: Becks Depression Inventory. In the third column, numbers in parentheses indicate degrees of freedom
Group comparisons indicated significantly lower ADΔhappiness-disgust scores in the group reporting reliable worsening at follow-up compared to the rest of the sample (d = 1.18) as well as a higher absolute BDI-II score at follow-up (d = 2.25) in the group characterized by reliable increase, and a trend-level effect indicating that individuals in the group reporting reliably increased symptoms had experienced more depressive episodes prior to study inclusion than individuals in the group not reporting reliably increased symptoms (d = 0.91).
Post-hoc analyses
Impact of Prior Depressive Episodes
We followed up on the latter observation by testing the association between the number of depressive episodes and ADΔhappiness-disgust in a separate linear regression model while controlling for participant age. This test did not yield a significant overall model fit either in the sample as a whole (F(2,60) = 0.76, p = 0.47, R2 = 0.02) or the subgroup of individuals with a history of major depression (F(2,32) = 0.24, p = 0.79, R2 = 0.01), and individual predictor estimates were not significant within the tested models. We did, however, observe an additional significant effect of the number of depressive episodes when adding this predictor into the a-priori logistic regression model described at the beginning of the previous section. In this alternative model, we predicted reliable worsening of symptoms from ADΔhappiness-disgust and the number of previous episodes while controlling for participant age and initial depressive symptoms. The overall model was statistically significant (R2 = 0.35, (4, 58) = 12.27, p = 0.02). Both ADΔhappiness-disgust (β =—0.07, p = 0.02) and number of episodes (β = 0.92, p = < 0.04) contributed significantly but in inverse direction, indicating that a higher ADΔhappiness-disgust score was less likely to cooccur with subsequent reliable worsening of depressive symptoms (OR = 0.94, 95% CI = [0.87–0.99]), while a higher number of depressive episodes prior to study inclusion increased the likelihood of experiencing reliable worsening (OR = 3.54, 95% CI = [1.06–6.02]).
The Role of Attention to Happiness in Relation to the ADΔHappiness-Disgust Differential Score
Descriptive statistics in Table 2 show that the mean difference in disengagement time between the observed groups (reliable worsening present | absent) at baseline was marginal for disengagement from disgusted faces (1.38 ms difference in means), but very pronounced for disengagement from happy faces (21.56 ms difference in means). While disengagement from happy faces appeared to be much faster in individuals subsequently experiencing reliably worsened depressive symptoms, this group difference was not statistically significant. Meanwhile, empirical data from other studies (e.g., Bartoszek & Winer, 2015; Salem et al., 2017) as well as theoretical contributions (e.g., Gallagher et al., 2024) have corroborated the notion that depressive symptoms are in fact associated with reduced attention to positive, potentially rewarding stimuli.
This preexisting connection prompted us to conduct a post-hoc analysis with the aim of clarifying whether the positive association between ADΔhappiness-disgust and reliably worsened depressive symptoms at follow-up (Hypothesis 2) might have been driven primarily by aberrant disengagement from happy faces. To this end, we constructed a series of alternative logistic regression models (parametrization summarized in Table S1, supplemental material). Firstly, we tested whether reliable worsening of depressive symptoms (present | absent) was predicted by attentional disengagement (AD) from happy faces alone, while keeping the model otherwise identical to the one described in the initial analysis, with lifetime history of depression and initial symptom severity entered as control variables. The same was then done for AD from disgust alone. Neither AD from happy faces nor AD from disgusted faces was a significant predictor on its own for later experiencing reliable worsening of depressive symptoms, either when entered into the model alone, or when each accompanied by the two previously used control variables (depressive symptom severity at baseline and lifetime history of depression). The overall model fit was insignificant in both cases. Next, individual mean disengagement latencies (AD from happy faces and AD from disgusted faces) were entered as two separate predictors into the same logistic regression model. Again, initial symptom severity and lifetime depression history served as control variables. When both AD from happiness and AD from disgust were entered into the model simultaneously, both contributed significantly to predict the presence or absence of a reliable increase of depressive symptoms at follow-up, and continued to do so when depressive symptom severity at baseline and lifetime depression history were controlled for. As one might expect, the predictive influence within the full model was in inverse direction for the two variables (negative for AD from happy faces and positive for AD from disgusted faces), so that relatively longer saccade latencies to disengage from disgust paired with relatively shorter saccade latencies to disengage from happiness were statistically predictive of reliable symptom worsening.
Receiver Operating Characteristic Curve Analysis
ROC analysis was performed to determine the area under the curve (AUC), sensitivity, and specificity when predicting the binary criterion of reliable worsening of depressive symptoms (reliable worsening present | not present) from the ADΔhappiness-disgust score alone (model 1). The estimated AUC for this model was 0.82, with a 95% CI of [0.71 – 0.93]. The maximum sensitivity achieved by this model was 1, with a corresponding specificity of 0.70. For this model, the difference score best classifying cases was –6.31 ms when using Youdens J to identify the optimal cutpoint on the ADΔhappiness-disgust difference score.
In model 2, the criterion of reliable worsening of depressive symptoms was predicted from the ADΔhappiness-disgust score, baseline depressive symptom severity and lifetime history of depression all entered simultaneously. The estimated AUC for this model was 0.92 with a 95% CI of [0.85 – 0.99]. Maximum sensitivity achieved by this model was 1, with a corresponding specificity of 0.88. Figure 2 depicts a visual comparison of both models.
Fig. 2
Receiver operating characteristic curve for binary classifier predicting reliable worsening of depressive symptoms (yes | no) from model 1 and model 2. In model 1, reliable worsening of depressive symptoms was predicted by ADΔhappiness-disgust alone (dark gray line). In model 2 reliable worsening was predicted by ADΔhappiness-disgust, depressive symptom severity at baseline and lifetime history of depression (light grey line). Dashed line serves as a reference to indicate chance level
×
Discussion
Discussion of Main Results
As its principal finding, the present investigation demonstrated that a smaller difference between time to disengage from happy and disgusted facial expressions (ADΔhappiness-disgust score) at initial assessment was positively associated with the likelihood of experiencing a reliable increase in depressive symptoms at 6-month follow-up. Importantly, this association remained significant when controlling for depressive symptoms at baseline and a lifetime history of major depression, the latter of which was in itself a significant independent predictor. In the binary classifying model fit to our sample’s data, a difference of 6 ms between the emotional conditions “disengagement from happy” and “disengagement from disgusted” (with saccade latency for disengagement from happy exceeding latency for disengagement from disgusted) emerged as the optimal cutpoint to distinguish individuals who experienced a significant increase in depressive symptoms at 6-month follow-up from those who did not. Notably, all individuals with a reliable increase in depressive symptoms showed an ADΔhappiness-disgust score below this cutpoint. Scores below the cutpoint hence predicted a reliable increase in depressive symptom severity with perfect sensitivity. Contrary to our primary hypothesis, we did not observe a significant association between the difference of disengagement latencies for happy and disgusted faces and dimensional change in depressive symptoms.
Research relating direct indices of gaze disengagement to future changes in depressive symptom severity is extremely sparse. To our knowledge, only two studies in the current body of literature have investigated this prospective association: Firstly, Sanchez-Lopez et al. (2019) investigated associations between gaze disengagement from disgusted as well as happy facial expressions and depressive symptom levels at 5-month post initial assessment in a non-clinical student sample. Gaze parameters and depressive symptom levels were assessed to test a more comprehensive moderated mediation hypothesis also including ruminative brooding (mediator) and adverse life events (moderator to the mediator). In that study, longer time to disengage from a positive facial stimulus at baseline indirectly predicted depressive symptom decrease at 5-month follow-up via its predictive influence on changes in brooding during the observation period (mediation); this predictive relationship in turn depended on the number of adverse life events experienced in the meantime (moderator to the mediated effect). In the present research design, we examined a different type of sample, and considered neither changes in ruminative brooding nor adverse life events. Nevertheless, there are some overlapping observations: Sanchez-Lopez et al. (2019) report that a direct dimensional association emerged neither between attentional disengagement from disgust and depressive symptoms five months later (r = 0.05) nor between attentional disengagement from happiness and later depressive symptoms (r = 0.08). This lack of dimensional associations between attentional disengagement parameters and later depressive symptoms could be interpreted to reflect consistency with our failure to find evidence of an association between dimensional change in depressive symptoms and the ADΔhappiness-disgust score (Hypothesis 1). However, because Sanchez et al. did not report on analyses testing the predictive power of differential processing between conditions, comparability of these results with our design remains somewhat limited.
In the second study relating direct indices of gaze disengagement to future aspects of depressive symptom severity, Yaroslavsky et al. (2019) analyzed effects of disengagement from sad faces and happy faces on current depression symptoms (cross-sectionally) and depressive affect ascertained over the next seven days. This study was done in adults with differing depression histories and varying current symptom levels, some of whom were experiencing a depressive episode at the time of assessment. Delayed disengagement from sad faces was found to predict elevated symptoms of depression cross-sectionally as well as a disadvantageous ratio of positive and negative affect over the next seven days, irrespective of previous history of depression. Because of the sample composition, this design did not allow to disentangle if the prospective associations observed could be explained by depressive symptoms present at the time, or whether they are indicative of a mood independent, more latent disposition that continues to undermine emotional wellbeing even after depression remits.
Beyond the somewhat narrow focus on disengagement of attention, there are to date two studies that have investigated prospective changes in symptom development as a function of attentional biases in the population of formerly depressed individuals. Firstly, a large prospective study reported null findings from a sample of initially remitted individuals tested at two- and four-years follow-up (Elgersma et al., 2019): Attentional bias for negative verbal stimuli in an exogenous cueing task relying on motor responses to inform bias indices neither showed prognostic value for recurrence into a new depressive episode, nor for depressive symptoms at either stage of follow-up. In addition, relatively low attentional bias for positive stimuli was not systematically associated with recurrence or change in symptom severity. In a second study, Newman et al. (2019) used eye tracking to assess dwell time on emotional facial expressions of sadness, threat (frightened / angry) and happiness in individuals in remission from major depressive disorder. Total dwell time for sad, threat, happy or neutral faces was not predictive of either depression symptoms or relapse into a new episode at 6-month follow-up. Again, this finding appears consistent with our failure to find evidence for Hypothesis 1. At the same time, it has to be noted that total dwell times comprise several attentional processing stages not included by the task we used to assess attention disengagement.
Taken together, studies to date have not provided evidence of a direct prospective association between attentional bias measures assessed in non-depressed states and the later development or recurrence of depression. Of note, most of the few existing studies differed from ours in sample selection, probing either analogous samples or patients with depressive symptoms present. In addition, some previous studies have used behavioral paradigms known to conflate attention measures with bias-irrelevant constructs, such as motor speed (for a discussion, see Rodebaugh et al., 2016). Our results are somewhat consistent with previous research in lacking evidence of an association between attentional bias measures and later dimensional changes in depressive symptom severity. However, the present prospective finding of an association between differential disengagement from disgusted and happy faces on the one hand and a reliable worsening of depressive symptoms on the other is novel.
Taking a critical stance towards this pattern of results, one may remark that the BDI-II has at times been found to underperform in comparison to other instruments in its sensitivity to detect change in depressive symptoms (Kounali et al., 2016; Titov et al., 2011). One could thus challenge the use of this measure for dimensional investigations altogether, arguing that it is generally ill-equipped to pick up on subtle to medium changes in depressive symptom severity. On the other hand, methodological follow-up investigations have found evidence to the contrary (de Beurs et al., 2019; Meesters et al., 2021), with the BDI-II’s sensitivity to detect change outperforming other measures. In addition, existing studies reporting on prospective associations between various cognitive biases and depressive symptoms have used the BDI-II for this purpose and indeed reported systematic dimensional associations across a timespan of several weeks or months (e.g. Everaert et al., 2015; Rude et al., 2002).
Adopting the stance that sufficient sensitivity of the BDI-II to reflect dimensional changes can be assumed, the present study’s dissociation between attention-based predictability of the dichotomous criterion of reliable worsening but not the continuous measure of dimensional change requires further consideration. It is important to note that the two endpoints chosen in the present study capture conceptually different things. In the case of Hypothesis 1, dimensionally relating within-sample variation of ADΔhappiness-disgust to variation in BDI-II score changes necessarily results in the inclusion of random intrapersonal changes on the BDI-II in the estimation of covariation between the predictor and the outcome variable. On the other hand, turning to Hypothesis 2, determining if the reliable change criterion has been met prior to fitting a logistic regression model is a way to assess if an intraindividual change exceeds the change to be expected simply as a consequence of the unreliability of the psychometric instrument in use. This method consequently “picks” or favors the portion of the observed variation in the outcome variable that is most likely not explained by measurement error. The latter is hence more likely to reflect pathogenetically meaningful changes in depression severity. Our use of this method may have allowed for the detection of an underlying longitudinal association between attentional bias and the clinically meaningful increase of depressive symptoms for the first time. With this in mind, we suggest the preliminary interpretation that the ADΔhappiness-disgust gaze parameter could be an attentional component well suited to specifically predict large and potentially meaningful increases in symptom severity. This notion is corroborated by the fact that we observed cases with reliably worsened depressive symptoms only in individuals who had experienced at least one episode of major depression prior to study inclusion, suggesting that the prospective association seen here is indeed reflective of a meaningful underlying pathomechanism.
Our findings bear a relationship to Reward Devaluation Theory (RDT; Gallagher et al., 2024; Winer & Salem, 2016), which posits that individuals with depression not only focus on negative stimuli but also actively avoid or inhibit engagement with positive and potentially rewarding material. In light of the fact that we observed a difference in means for disengagement from happiness in the specific sample of this study – though not significant on the level of inferential statistics – it seemed reasonable to investigate a posteriori if fast disengagement from happiness, more so than disengagement from disgust, had been driving the prediction observed for reliable symptom worsening in our initial model. Indeed, short saccade latencies to disengage from happiness were associated with a reliable increase of depressive symptoms, but only when entered into the model simultaneously with AD from disgust. This finding speaks to the specificity of ADΔhappiness-disgust as a differential processing score in predicting depressive symptoms.
Strengths and Limitations of the Present Study
This study is one of the very few works in clinically diagnosed samples relating oculomotoric parameters to a symptom outcome measure assessed several months later. It is the first investigation to show that aberrant attentional processing in remitted MDD predicts worsened depressive symptoms on the individual level over a period of several months. Our findings thus add important new information to the existing body of research, allowing us to evaluate the predictive value of attentional aberrations previously described in the context of remitted depression. In addition, the current investigation used an innovative eye-movement paradigm well suited to observe process-pure measures of attentional components unbiased by motor speed. As a result of the testing of two informative endpoints – AD’s association with dimensional change and AD’s association with reliable increase – we have been able to derive the preliminary conclusion that ADΔhappiness-disgust may be a differential marker suited to identify remitted individuals with an elevated risk of experiencing reliable symptom worsening, and possibly with a greater need for preventive care in general. Future replication provided, our finding may be used to generate ideas for projects probing the malleability of attentional aberrations in remitted MDD.
Several limitations of our study are equally worth noting. Firstly, this being an exploratory follow-up study, its sample size was limited by the number of participants in a previous cross-sectional study. Limited sample sizes decrease the power for detecting prediction effects, and also pose the risk of overestimating effect sizes. At the current stage of inquiry, the significant association between the ADΔhappiness-disgust score and a reliable worsening of depressive symptoms must be interpreted with caution. Nevertheless, it seems remarkable that the ADΔhappiness-disgust score predicted reliable worsening of depressive symptom severity with perfect sensitivity in this sample. Given the scarcity of longitudinal designs relating bias measures to symptom outcomes across clinically meaningful timespans, we suggest that the pattern of effects observed here should be seen as a promising preliminary finding and spur further investigation. In a similar vein, due to limited variation within this predictor variable in our sample, we could not draw definitive conclusions about the association of the ADΔhappiness-disgust score with depression recurrence, i.e., the frequency and absolute number of past depressive episodes. Gaining more data suited to assess this connection would certainly add important information to this line of inquiry both clinically and mechanistically.
Secondly, with the outcome measures chosen (e.g., BDI-II increase from baseline to follow-up ≥ 9), we adopted a parsimonious and easily replicable criterion for the assessment of the development of symptoms over time, but we did not assess relapse/recurrence, that is, if criteria of a depressive episode were met at follow-up that had previously been absent. Clinician-rated in-person assessments were conducted at the time of initial eye-tracking assessment, but we relied solely on the individual’s self-report during follow-up. Furthermore, the presence of depressive symptoms was assessed only once after a fixed time period of six months had elapsed, and we did not include retrospective reports of changes that might have happened earlier during the observation period. Thus, some changes might have gone undetected as a result of our study design, and multi-wave follow-up investigations may address this limitation.
More substantially, critics of sum-score measures widely used to assess changes in depression severity have reasoned that the heterogeneity and incommensurability of symptoms within the depressive syndrome inevitably leads to a blurring of possibly meaningful associative relationships when measures such as the BDI-II are employed. It has been proclaimed that the widespread use of sum-score measures in depression research contributes to the lack of progress in key research areas, and we should instead “analyze each symptom separately” (Fried & Nesse, 2015). This has not been done in the current preliminary study. Looking ahead, our procedure could be optimized by assessing symptoms of depression that may not even constitute MDE criteria, but are conceptually linked to the stimulus material presented in this experimental paradigm, such as perceived experiences of social isolation, self loathing, targeted rejection, and loneliness. Given that we did not assess specific dimensions within the multidimensional phenomenology of depression prospectively, we may have missed meaningful symptom-specific associations with experimental parameters.
Finally, aiming to assess potential aberrations in attentional disengagement unbiased by acute pharmacological effects, we have opted to only include individuals who were not taking antidepressant medication. With this in mind, the results presented here are best understood as a description of mechanistic associations under naturalistic conditions, and applicability to clinical samples including individuals that have received pharmacological treatment would have to be tested separately. In a similar vein, we did not systematically assess if participants initiated other forms of treatment, such as individual therapy, during the observation period that may have modulated the relationship between differential processing and subsequent changes in depressive symptoms.
Future Directions
To advance clinical application, we suggest that future studies should test (i) if the current observation is replicable, (ii) if changes in the ADΔhappiness-disgust disengagement pattern can be acquired and solidified through attention training and (iii) if this form of intervention results in the expected beneficial effects in the sense that it reduces the odds of increases in depressive symptom load. From what we have observed in the current study, a desirable bias modification could be achieved either by increasing speed for disengagement from faces signaling social disdain, or by training to slow down disengagement from positive facial expressions, or a combination of both, with the aim of shaping a pattern resembling observations in healthy, possibly resilient individuals. Existing trainings designed to modify attentional disengagement in depression-prone individuals have shown some promising results, but applications to date stem solely from analogue samples without characterization of lifetime depression status (Basanovic et al., 2021; Godara et al., 2019, 2021; Möbius et al., 2018; Sanchez-Lopez et al., 2021). Our results provide encouragement to further test this approach as a potential treatment avenue in currently non-symptomatic individuals previously affected by major depression. As recent work has demonstrated the potential to accelerate gaze disengagement by way of anodal tDCS over the left DLPFC (and slow it down, for that matter, by stimulating over the right DLPFC, Sanchez-Lopez et al., 2018), such additions may provide even further experimental opportunities to intensify training effects in individuals previously affected by major depression. Finally, future research should investigate if the association we observed between ADΔhappiness-disgust and reliably increased symptoms translates to individuals with mild to moderate residual symptoms of depression at initial testing.
Conclusion
Our findings support the theoretical assumption that difficulty in shifting attention away from emotionally negative material is associated with the development of depressive symptoms over time. ADΔhappiness-disgust might be a marker suited to predict increases in symptom load and may be helpful in selecting individuals who are at high risk for those impactful changes. Given the fact that we observed cases with reliable symptom gain only in individuals who had reported episodes of major depression prior to entering the study, the population of remitted individuals seems particularly well-suited to focus on when testing interventions designed to change disadvantageous patterns of attentional disengagement.
Acknowledgements
The authors express their appreciation to Lina Specht, Sarah Sundermeier and Franziska Jüres for their support in conducting this study.
Declarations
Competing Interest
There are no conflicts of interest to declare for each contributing author.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
share
DELEN
Deel dit onderdeel of sectie (kopieer de link)
Optie A:
Klik op de rechtermuisknop op de link en selecteer de optie “linkadres kopiëren”
Met BSL Psychologie Totaal blijf je als professional steeds op de hoogte van de nieuwste ontwikkelingen binnen jouw vak. Met het online abonnement heb je toegang tot een groot aantal boeken, protocollen, vaktijdschriften en e-learnings op het gebied van psychologie en psychiatrie. Zo kun je op je gemak en wanneer het jou het beste uitkomt verdiepen in jouw vakgebied.
Differential Gaze Disengagement from Friendly and Disdainful Facial Expressions Predicts Reliable Increase of Depressive Symptoms Following Euthymia: Preliminary Evidence from an Eye-Tracking Study at 6-Month Follow-up