Biased Maintenance of Attention on Sad Faces in Clinically Depressed Youth: An Eye-Tracking Study

Cognitive models of depression consider cognitive biases a key element in development and maintenance of MD [7, 8]. Cognitive biases describe a shift in three aspects of information processing: attention to, interpretation of and memory for emotional information [9‐11]. This shift promotes the filtering of incoming information according to the depressive schemas that initially activated the biases and in turn results in confirming these schemas, facilitating negative mood [8]. The shift in attention towards negative compared to positive or neutral information is referred to as negative attentional biases (AB). According to the model of vulnerability to depression by De Raedt and Koster [12], attentional processes are influenced by genetic and neuroregulatory factors [e.g. hypothalamic–pituitary–adrenal (HPA) axis functioning and impaired attentional networks] and are bi-directionally associated with emotion regulation (e.g. rumination). In this framework, negative AB would be essential in linking psychological and biological vulnerabilities for depression and its recurrence [12].

Importantly, attention is not a unitary process but comprises multiple consecutive components: orientation, maintenance and detachment of attention [13]. Studies on adult depression find reliable evidence for AB in the maintenance of attention on negative disorder-specific stimuli, while results regarding orientation of attention are less robust [11, 13]. Since major development occurs in cognitive and emotional functioning [14] and in attentional networks [15] during childhood and adolescence, results from adult studies cannot be directly transferred to youth. Cognitive patterns might not evolve into stable, trait-like “cognitive styles” until adulthood [16], therefore these might have less influence on cognitive vulnerability in youth than in adult depression. Alternately, brain maturation and hormonal changes associated with an enhanced emotional sensitivity [17] might make youth more susceptible to negative cues in ambiguous emotional information, resulting in more pronounced negative cognitive biases.

There is a growing body of literature on AB in depressed youth, yet findings are heterogeneous [18]: while some studies found youth with MD to show increased attention towards negative disorder-specific (sad) information [19, 20], another study found them to avoid sad stimuli [21]. An association between negative AB towards threatening information (angry faces) and depressive symptoms in unselected youth was observed by one study [22], while no evidence for AB was discovered at all in a different study [23]. These differences could be due to moderating factors like genetic differences in e.g. hypothalamic–pituitary–adrenal (HPA) axis reactivity [24], or due to the different methods employed or different aspects of attention measured (e.g. allocation, maintenance with or without distraction) in different samples. As such, the direction of AB, i.e. whether depressed individuals prefer or avoid (i.e., show AB towards or away from) negative stimuli is unclear. Results are also unequivocal regarding specificity, i.e. whether AB apply to disorder-specific stimuli (e.g. sad faces) or threatening stimuli (e.g. angry faces) or negative information in general.

To assess AB, reaction-time (RT) based behavioural measures are used regularly, with speed of response acting as indicator for AB towards or away from emotional stimuli. Investigating AB by measuring RT involves drawing indirect inferences about the focus of attention at one distinct point in time, neglecting that attention consists of different consecutive stages [13]. In addition, RT is influenced by non-attention-related processes like the speed of motor response, which may be generally delayed in MD [12]. The most frequently used paradigm, the Dot-Probe Task (DPT) [25], in which a probe appears in the location of either an emotional or a neutral stimulus, has been shown to have unacceptable psychometric properties [26‐28]. The emotional Visual Search Task (VST) [29], an alternative paradigm measuring RTs to identify a target emotional stimulus among distractors, shows better reliability [30, 31].

A more direct approach to assess attention is registering overt visual attention by tracking eye movements [13]. In eye-tracking (ET) studies, free viewing paradigms like the Passive Viewing Task (PVT) [21] are used most frequently. Participants are presented with various emotional stimuli to look at without further instruction. This method is not influenced by (non-ocular) motor response and, more importantly, provides data of visual attention over a longer period of time, enabling to distinguish between the various components of visual attention (unlike RT based measures). As Högström et al. [32] elaborate, the vigilance-avoidance-hypothesis in research of AB in anxiety states that differentiating categorically between visual attention towards (vigilance) or away from (avoidance) salient stimuli might not grasp the attentional process accurately. Instead, there is evidence from eye-tracking research that individuals with anxiety show initial vigilance to disorder-related stimuli, followed by avoidance (see e.g. [32‐34]). As both, initial orientation of attention as well as maintenance of attention can be assessed via the PVT, it allows to investigate if the vigilance-avoidance-hypothesis or similar more complex attentional mechanisms also apply to youth depression.

Additionally, psychometric properties of the PVT have been found acceptable in adults [27, 35] and in youth [36]. To our knowledge, only two studies have employed this paradigm to investigate bias in sustained attention in youth depression: one found AB away from sad faces in youth with MD [21], the other found AB towards sad faces in youth at risk for MD, moderated by genetical differences in HPA-axis reactivity [24].

We compared 9–14-year-olds¹ with MD to two groups of non-depressed children and adolescents varying in their risk for depression: children of parents with a history of depression, known to have an increased risk for MD themselves (high-risk; HR), see e.g. [39] and children of parents with no history of any mental disorder with a low risk for depressive disorders (low risk; LR). The comparison of those three groups enabled us to pursue the study’s second aim: to determine whether AB are more pronounced in currently depressed youth compared to youth at elevated risk for depression. HR youth have been found to show AB for negative information, although their direction and specificity has not been established yet with evidence for AB towards sad [40, 41] or angry [42] stimuli, as well as for AB away from sad stimuli [43]. Negative AB in children and adolescents at risk for depression would indicate that AB might act as a cognitive vulnerability contributing to the development of depression, as suggested by theoretical models, e.g. [7, 8]. Even more pronounced AB in currently depressed children and adolescents however would indicate that these biases might be exacerbated as a consequence of depressive symptomatology.

No study to date has directly compared AB in depressed, HR and LR youth, however, studies on interpretation [44] and memory [45] biases have found negative biases to be present in both depressed as well as high-risk youth compared to low-risk youth but to be more pronounced in currently depressed youth than in high-risk groups.

With respect to our first aim we expected to find more negative AB in currently depressed youth compared to healthy youth (both high- and low-risk), in accordance with theoretical models [8] and previous research [18]. Regarding specificity, we expected AB for sad faces (i.e. mood-congruent depression-specific stimuli) rather than angry faces (i.e. threatening anxiety-specific stimuli) in both the MD and HR groups. Due to differences in direction of AB in previous studies, we did not make predictions on direction of AB, i.e., whether MD participants would look preferably towards negative stimuli or avoid them. With respect to our second aim, we expected negative AB to be present in youth at high risk for depression compared to youth at low risk for depression (according to [40‐42, 46]) but to be more pronounced in the MD versus HR group, as found for interpretation and memory biases [44, 45].

In total, 124 children and adolescents aged 9–14 years were included in data analysis.³ Participants formed three groups based on their current mental health state and risk of developing MD: n = 32 youth with MD (MD group), n = 49 mentally healthy youth at high familial risk for MD (HR group) and n = 43 mentally healthy youth with low risk for MD (LR group). Sample size was based on an a priori power analysis (α error probability = 0.05; power = 0.8; two-tailed) for the comparison of HR and LR groups (a smaller effect size was expected for this effect than for comparison with the MD group). Based on effect sizes (ES) from previous studies with mood induction procedures [40, 41], we expected an ES of at least d = 0.7, resulting in a required sample of n = 34 per group.

Participants with MD were recruited mainly at a University Hospital for Child and Adolescent Psychiatry in Germany, some by a cooperating licensed psychotherapist and some responded to public adverts. Data of the LR and HR groups was largely (87%) derived from a study evaluating the transgenerational transmission of cognitive biases [36, 38] in which children and adolescents participated with one of their parents. N = 29 HR participants were recruited from a previous study on a preventive intervention for this particular risk group [48], n = 12 of them had taken part in the programme by the time they were tested in the present study. Others, as well as children in the LR group, were recruited by local and online advertisements and by letters to randomly selected families with children in the corresponding age range provided by the local registry office.

Youth were included in the MD group if they currently met criteria for MD according to DSM-IV [49]. Of those 32 youths, n = 4 fulfilled criteria for a recurrent course of MD, n = 2 were in partial remission, n = 15 had at least one comorbid anxiety disorder and n = 3 were treated pharmacologically. The HR group comprised youth who did not meet criteria for any axis I disorder, current or past,⁴ but at least one of their parents fulfilled diagnostic criteria for MD (n = 47; of which n = 8 fulfilled diagnostic criteria at the time of participation) or dysthymia (n = 2) during the child’s lifetime. Youth were not included if their parents had a history of bipolar disorder, schizophrenia or substance abuse. Participants were included in the LR group if they did not meet criteria for any axis I disorder currently or in the past and none of their parents met criteria for any axis I disorder at any time. All participants had normal or corrected to normal vision. Exclusion criteria for participants in all three groups were an intelligence quotient (IQ) < 85,⁵ assessed with the CFT-20-R [50], a pervasive developmental disorder, attention deficit and hyperactivity disorder or a history of schizophrenia or bipolar disorder.

All procedures were approved by the ethics committee of the Medical Faculty of the LMU Munich. Written informed consent was obtained from all participants and parents after a comprehensive explanation of the experimental procedures. HR and LR youth who participated with their parents in the previous study on transgenerational transmission of cognitive biases received a reimbursement of €50 while MD participants who took part without their parents only in this study received €30.

Before inclusion, all participants underwent extensive diagnostic assessment. The K-DIPS (“Diagnostisches Interview bei psychischen Störungen im Kindes- und Jugendalter”) [51], a standardised, semi-structured psychiatric diagnostic interview for assessment of psychiatric diagnoses in youth, was administered to participants and one of their parents. The K-DIPS is a well-established German diagnostic measure based on diagnostic criteria of axis I disorders according to DSM-IV [50] with good interrater-reliabilities [52]. Interviews were conducted and evaluated by trained interviewers. Interrater-reliability was determined for 18% of HR and LR participants and revealed an accordance rate of 100% for lifetime diagnosis of depression (pre-defined criterion). Interviews from the MD group were not re-rated, but most MD participants were receiving treatment for a clinical diagnosis of depression that was confirmed with the diagnostic interview. Psychiatric diagnoses of LR and HR participants’ parents were assessed with the adult version of the interview, the Diagnostisches Interview bei psychischen Störungen (DIPS), [53], with good interrater-reliability [54]. Accordance rate for lifetime diagnosis of depression was 94% in our sample. In the HR group, the DIPS was administered only to the parent affected by depression; in the LR group the DIPS was administered at least to one parent and whenever possible to both parents.⁶ The German version of the Beck Depression Inventory-II [55], obtained from both parents for 83% of LR youth, was used to measure parents’ depressive symptoms. Scores differed significantly (t_64.5 = 6.0, p < 0.001) between parents of HR (M = 9.9; SD = 8.5) and LR participants (M = 1.9; SD = 3.5).

For the assessment of the participants’ depressive symptoms the German version of the Children’s Depression Inventory “DIKJ” [56] was used and symptoms of anxiety were measured by the trait scale of the German version of the State Trait Anxiety Inventory for Children, STAIC [57]. For 123 of 124 included participants, a score for depressive symptoms could be determined and for 119 of 124 children an anxiety score was available. In our sample, reliability of both self-report measures was excellent (DIKJ: Cronbach’s α = 0.96; STAIC-T: Cronbach’s α = 0.92).

Coloured photographs of children’s faces displaying sad, angry, happy and neutral expressions on black background served as stimuli in both tasks [58]. Half of the models were male, half female in each task. The stimulus set comprised photographs of 16 models with sad, angry and happy expressions in the VST and 24 models with sad, angry, happy and neutral expressions in the PVT.

An emotional VST, adapted from [29] was used to assess AB in an RT-based task with explicit instructions to attend to specific emotions. The task was administered during ET, but due to poor reliability of the ET indices, ET results are reported in the supplement.

Each trial started with a drift correction (small white circle in the screen’s centre). Upon fixation of the circle, the trial was initiated. A fixation cross was then presented for 500 ms in the screen’s centre. Subsequently, the stimuli were presented in a 4 × 4 grid containing 15 distractors and one target (see Fig. 1). Each stimulus display contained all 16 models randomly allocated to the 16 grid positions. The participants’ task was to identify the face showing a certain emotion (i.e., the target) as quickly as possible and click on it with the mouse. Time to identify the target was unlimited. The experiment consisted of four blocks: Two blocks with happy targets and negative (either sad or angry) distractors and two with negative targets and happy distractors. Each block comprised 32 trials with the target appearing twice in each position and being twice each model. Blocks as well as trials within the blocks appeared in random order. Before each block, participants completed three practice trials.

As a purely ET-based measure of AB, enabling participants to freely gaze at the emotional stimuli, a PVT was administered [21]. Each trial began with a drift correction (see above). A fixation cross followed for 1000 ms. Then the 2 × 2 stimulus array was presented for 15000 ms. The task consisted of 16 emotional trials, corresponding to the minimum trial number suggested for ET research [61] and eight neutral trials (not analysed), presented in random order. In the emotional trials, each array consisted of four photographs of the same model displaying a sad, angry, happy and neutral facial expression (see Fig. 2). The position of each emotional facial expression was randomly assigned to one of the quadrants with each emotion being presented four times in each quadrant. The neutral trials comprised four photographs of the same model with a neutral facial expression. Stimuli were sized approximately 9.5 cm × 7.5 cm and were presented with a distance of approximately 6.5 cm horizontally and 1 cm vertically between. Participants were instructed to fixate the white circle and fixation cross and then freely view the stimuli, required only to keep their attention focussed on the screen.

The course of the experimental session corresponds to [38]. Tasks were presented in random order and a mood induction procedure was administered twice during the experimental session. Participants watched a 2 min scene from the movie The Lion King [64] that had successfully induced unpleasant mood in children and adolescents before [65]. In our study, participants in both healthy groups reported a significantly worse mood, assessed using the valence dimension of the 9-point Self-Assessment Mannequin scale [66] after watching the movie scene compared to baseline (Fs ≤ 43.22; ps < 0.001). Mood did not change significantly, however, in the MD group (F2.4 = 1.15; p = 0.33). Details are presented in Supplement 5.

Statistical data analysis was conducted with SPSS25. Significance level was set to p = 0.05 (two-tailed).

For the VST, one-way analyses of variance (ANOVAs) were conducted to assess group differences in AB_VST scores. To assess relationships between psychopathology and AB, correlations were calculated of AB_VST score with depression and anxiety scores.

For the PVT, a repeated-measures ANOVA with within-subject factors Time Window (5: time windows 1–5) and Emotion (4: angry, happy, neutral, sad) and the between-subjects factor Group (3: MD, HR, LR) were performed for percentages of dwell time. Degrees of freedom were adjusted via Greenhouse–Geisser correction when necessary. As this study’s main focus was a group comparison, only significant effects involving the factor Group were followed up using post-hoc ANOVAs and t-tests. To assess relationships with psychopathology, correlations of dwell time on each emotion (averaged across time windows) with depression and anxiety symptoms were computed.

All analyses were repeated excluding participants with psychopharmacological treatment (n = 3), as this has been found to reduce negative cognitive biases [67] and enhance attention toward positive stimuli [68]. The pattern of results in our sample remained the same, so findings are reported based on the whole sample. We did not include age as a covariate in our analyses as this is not appropriate when the between-groups factor (group) and covariate (age) are non-independent [69]. However, we performed additional analyses to investigate if age may have accounted for our results (see below).

Sample characteristics are presented in Table 1. Groups did not differ in gender ratio or IQ but in age: participants in the MD group were significantly older than participants of the HR and LR groups. As expected, the groups also differed in psychopathology: the MD group reported higher depression and anxiety symptoms than the healthy groups, which did not differ from each other, indicating that the HR group was as psychiatrically healthy as the LR group.

One-way ANOVAs on AB_VST scores revealed no significant differences between MD, HR and LR youth (Fs < 1; ps > 0.1) regarding sad or angry faces. The AB_VST for sad faces was significantly < 0 in the LR group (t₄₁ = 2.5, p = 0.016), indicating a positive AB for sad faces (i.e. away from sad faces/towards happy faces), but bias scores for sad faces in the other groups as well as for angry faces were not significantly different from 0 (ts ≤ 1.6, ps > 0.1). Bias scores are presented in Table 2. The correlational analyses revealed no relationship between AB_VST scores for sad or angry faces with depressive or anxiety symptoms (r ≤ 0.17; p ≥ 0.072).⁸

The TimeWindow × Emotion × Group ANOVA revealed a significant main effect of TimeWindow (F_1.9,207.3 = 24.3, p < 0.001, η_p² = 0.18) and Emotion (F_2.6,295.1 = 4.5, p = 0.006, η_p² = 0.04). Significant TimeWindow × Emotion (F_9.6,1076.0 = 1.9, p = 0.040, η_p² = 0.02) and Emotion × Group (F_5.3,295.1 = 2.5, p = 0.027, η_p² = 0.04) interactions emerged. All other main effects and interactions were non-significant (Fs ≤ 2.3; ps ≥ 0.1). To follow up the Emotion × Group interaction, ANOVAs with the factor Group were calculated separately for each emotion on percentage of dwell time averaged across time windows: a significant effect emerged of Group for sad faces (F_2,112 = 4.7, p = 0.011, η_p² = 0.08), but no other emotion (Fs ≤ 2.5, ps ≥ 0.08). Post-hoc t-tests revealed a significant difference in percentage of dwell time on sad faces between the MD and HR groups (t_36.2 = 2.3, p = 0.027, d = 0.6) with the MD group dwelling more on sad faces than the HR group. Differences between MD vs. LR groups (t_33.4 = 1.5, p = 0.145, d = 0.4) and between HR and LR groups (t₈₅ = 1.7, p = 0.090, d = 0.4) were non-significant, with a trend for HR participants dwelling less on sad faces than LR participants. Percentages of dwell time are presented in Fig. 3. Furthermore, positive correlations of dwell time on sad faces with depressive (r = 0.29, p = 0.002) and anxiety symptoms (r = 0.38, p < 0.001) and a small negative correlation between dwell time on happy faces and depressive symptoms (r = -0.21 p = 0.023) were found among all participants.⁹

As groups differed in age, we performed additional analyses to investigate if this may have accounted for the group differences in dwell times on sad faces. Pearson’s correlations between age and dwell time on sad faces were computed separately within each group (to rule out that correlations were artifacts of group differences): no significant correlations emerged (rs ≤ 0.23, ps > 0.1).

To our knowledge, this is the first study directly comparing AB in currently depressed, high-and low-risk youth. It contributes to a more detailed understanding of AB in youth depression and their role in development and maintenance of MD. Another strength is our use of a standardised, semi-structured diagnostic interview with all participants, conducted by trained personnel to ensure diagnostic accuracy. Moreover, parental psychopathology was also assessed via a psychiatric interview in the HR and LR groups rather than relying on self-report measures. In addition, we used multiple measures of AB, assessing attention in deliberate vs. externally guided circumstances. We evaluated reliabilities of our measures, a practice often omitted in AB research [36], and only interpreted the results of those measures with at least acceptable reliability. This led to differing results, contributing to a more nuanced understanding of AB in youth depression.

With most MD participants recruited at a Department of Child and Adolescent Psychiatry or through outpatient psychotherapists, most had received psychotherapy at the time they participated. Unfortunately, we did not systematically assess treatment type or duration. As psychotherapy is expected to reduce AB (e.g. in anxiety disorders; see e.g. [78‐80]) effect sizes might be underestimated in our study. Furthermore, we recruited the HR group partly from a study evaluating a prevention program for at-risk youth [49] some had already participated in. This Cognitive Behavioural Therapy based group intervention conveyed distraction as a coping strategy, among others, which might have influenced participants’ volitional attention maintenance and added to our HR group’s resilience, rendering them not entirely representative for children of depressed parents. Another limitation results from comorbid anxiety disorders in nearly half of the participants in the MD group, known to be associated with AB, as well [80]. However, anxious youth are found to show AB for threatening stimuli [80‐82] while depressed individuals are commonly found to show AB for sad stimuli [13]. Since we found AB for sad faces only, it is more likely that the AB we captured were related to participants’ depression rather than the comorbid anxiety disorders. In addition, our three groups differed significantly in age with the MD group being the oldest. As the incidence of MD increases with puberty [83] this does not come as a surprise. However, since the detected AB did not correlate with age, it is unlikely that this difference accounted for our results. Furthermore, it has to be emphasized that although our sample size is larger than previous studies’ (e.g., [21]: n = 19 in the MD group; [23]: n = 19 in the MD group; [73]: n = 10 in the high-risk group), the sample size was only powered to detect medium ES and may have prevented us from finding smaller effects. At the same time, it is possible that the significant effect we found is not a real group difference but a false positive result since we did not correct for multiple testing (in line with, e.g. [21]). Replication in larger samples of children and adolescents is therefore crucial to rule out that our result is a chance finding and to be able to draw firmer conclusions.

Finally, even though we determined and reported reliabilities of our measures and the correlation of the PVT maintenance score with depressive symptoms indicates that it could be a valid indicator of depressive processing, we cannot provide a full description of the psychometric properties of our tasks. The appropriateness of measures that have often been used to assess AB in the past, like the DPT, has repeatedly been called into question [26‐28], while at the same time no alternative measure has yet become the widely accepted norm. Therefore, future studies should further examine reliability and validity of tasks assessing ABs, refine existing tasks and develop new ones that can measure AB with less error. As this should reduce the variability of results, it would truly move the field forward.

In summary, our study provides preliminary evidence for a more negative AB in depressed youth compared to healthy youth at risk for depression. Our sample of youth with MD preferentially attended to sad faces, i.e., their AB was directed towards negative information. This AB was found only regarding sad, not angry faces, suggesting AB in youth depression may be specific to disorder-related negative emotional information. In addition, the more negative AB in depressed youth was found only for maintenance of attention in a task where attention was allocated deliberately, suggesting this AB might be specific to an internally controlled and at least to some extent conscious aspect of attention. Our results suggest that depressed and at-risk youth might show AB in opposite directions with AB towards negative information in depressed youth and AB away from negative information in at-risk youth, potentially acting as a protective mechanism in this group. However, this interpretation remains highly speculative and replication in larger samples is essential. Longitudinal research is necessary to address the temporal course of AB in youth, i.e. if youth with AB away from negative information are less likely to develop a depressive disorder than those with negative AB towards or if negative AB arises with depressive symptomatology. This will help shed more light on the role negative AB play in the development and maintenance of depressive symptoms. We think that despite the speculative nature of our interpretations they provide an interesting incentive for the elaboration of theoretical frameworks and starting points for future investigations.