Introduction

A large body of research has demonstrated attentional biases when viewing emotional stimuli in individuals with emotional disorders. Eye tracking technology provides a direct and continuous measure of overt visual attention, and has been used as an important metric in emotion recognition (ER) tasks. In a meta-analytic review of eye-tracking and affective disorders, results showed that when presented with an array of at least two stimuli, depressed individuals were characterized by reduced orienting to positive stimuli, reduced maintenance of gaze on positive stimuli, and increased gaze on dysphoric stimuli [1, 2]. Individuals with depression may voluntarily gaze less at positive stimuli because they are less sensitive to the pleasantness of it, reducing their incentive to maintain gaze. Research has suggested that the strong anhedonic bias in depression is not unique to depression, but an aspect of low positive affect more generally [1, 3, 4].

The relationship between anxiety-related processes and visual attention has been the subject of many research studies (e.g., [5–7]), but there is still not a consensus on how different types of anxiety symptoms and their severity impact visual attention. In general, individuals with anxiety disorders show increased vigilance to threat during free viewing and visual search and difficulty disengaging from threat during visual search but not during free viewing tasks relative to controls [1]. Recent work has examined the relations between worry, rumination, and visual attention, and found that self-reported rumination is linked to greater attention to sad stimuli [8, 9] but worry, compared with rumination, leads to relative avoidance of positive information [10]. Whether this is true across anxiety disorders, representing a transdiagnostic process, remains to be tested.

Relative to common anxiety disorders like generalized anxiety disorder, more is known about visual attention in social anxiety disorder, where eye gaze is likely to be more relevant to the maintenance of symptoms (e.g., [11]; see Staugaard [12] for a review). Individuals with social anxiety disorder may be more sensitive to detecting threat, and may misinterpret faces as threatening when they make quick judgments [13, 14]. In a dot-probe task with emotional faces, Schofield and colleagues [14] found that social anxiety was associated with attention to emotional (rather than neutral) faces over time, and difficulty engaging attention from angry faces. Heuer and colleagues [13] administered a morphed faces task (a task involving watching a series of computer-morphed faces that change slowly from a neutral to a fully emotional expression) to socially anxious individuals and non-anxious controls with time pressure (restricted viewing task, RVT) or with unlimited viewing of the faces (free viewing task, FVT). Participants with high levels of social anxiety demonstrated a threat bias (disgust interpreted as contempt) in the RVT, contrasting with the non-anxious control group’s positive bias (disgust interpreted as happy). No group differences were found in the FVT. Thus, time spent on faces and the pressure of making an accurate judgment in ER tasks may be linked with levels of anxiety and social anxiety. Additionally, more recent research has indicated that performance-based social anxiety specifically, when compared to panic disorder with agoraphobia, appears to present a sustained bias for vigilant attention to aversive facial expressions, but for the most severe social anxiety, patients show an opposing avoidance of aversive facial expressions [15]. This indicates that the severity as well as type of social anxiety play a role in emotional face perception and visual attention.

While much of the literature has looked at the eye region as an area of interest in ER some researchers have defined areas in the lower face (mouth) as particularly important in recognizing emotion. It is important to consider areas of interest within the entire face, not just the eye region, when examining emotion recognition in clinical groups, as there may be differences in visual attention patterns based on disorder status or severity. For example, in highly socially anxious individuals, looking at the mouth may be more comforting than staring into eyes. Wong and colleagues [16] examined patterns of visual scanning as predictors of emotion identification in older adults and younger adults. Older adults who made more fixations in the top halves of faces were more accurate at identifying emotions than those who made more fixations in the lower halves of faces. However, for the emotion of disgust in particular, older adult participants were more accurate when they fixated on the lower half of the face. In the younger adult group, analyses revealed a significant positive correlation between emotion-identification accuracy and difference scores computed on the number of fixations made to the top half versus bottom halves of faces [16]. In another study, individuals with high-functioning autism fixated more in the mouth region of faces even when the faces were inverted [17]). These authors suggest that abnormal gaze in individuals with high-functioning autism is driven by an impaired top-down strategy for allocating visual attention. Accordingly, it seems necessary to examine eye gaze patterns for each of the emotional faces when gauging where participants are attending.

The current study examines clinical factors as dimensional variables within a patient sample that is highly comorbid on depression and anxiety diagnoses. While prior research has compared visual attention patterns in depressed and anxious groups to healthy controls, the comparison of visual attention patterns in clinical groups is lacking. Moreover, while the emotion recognition literature has established impairments in emotion recognition in samples with anxiety and mood disorders compared to the general population, we do not yet know if a reason for these deficits is based in differences in time spent on certain regions of interest on an emotional face (eyes vs. mouth) or due to other reasons. Previous group-based comparisons may have led to inconsistent or weaker ER results because individual differences in symptom severity were collapsed into a binary variable. Our approach takes into consideration the heterogeneity of depression symptoms, the severity of anxiety symptoms, and their comorbidity [18]. In addition, we were equipped to attempt to disentangle social anxiety factors from other forms of anxiety.

In order to clarify the relationship between visual attention and anxiety and mood symptom severity, we used an entirely clinical sample with anxiety and mood disorder diagnoses. We had four hypotheses, each based on prior research. First, we expected there would be more fixations on negative faces in depressed participants compared to anxious based on the prior finding that depressed individuals attend more to dysphoric stimuli (e.g., [1]). Of note, this hypothesis was based on a finding comparing depressed individuals to healthy control subjects, and it remains to be tested to see if the anxiety group differs significantly from the depressed group. Next, we hypothesized that higher levels of depression would be associated with more fixations on sad faces based on the established finding that depressed individuals show a specific attentional bias to sad, but not angry or threatening facial expressions (e.g., [8, 19]). Third, we predicted that higher anxiety scores would predict more fixations in threatening faces (fear) based on the prior finding that social anxiety is associated with difficulty disengaging from threat (e.g., [14, 20]), specifically the indirect or ambiguous threat of a fearful face. Of note, we used an emotion recognition task that was different from the findings for which we based our prediction, as many tests of threat-bias include a paired face forced choice paradigm, and we used a dynamic emotional face task. Additionally, there is a body of work that shows that angry faces can elicit a threat-related attentional bias in anxious individuals (see Bar-Haim et al, 2007) [21], so we examined visual attention differences in angry faces (direct threat) in addition to fearful faces (indirect threat) based on level of anxiety. Last, we wanted to explore if there was reduced fixation on positive faces in high levels of worry and anxiety. Based on the finding that worry, compared to rumination, leads to relative avoidance of positive information ([10] we expected that higher anxiety severity would predict reduced fixation on happy faces.

Materials and methods

Eye tracking

Hardware.

An Applied Science Laboratories Eye-Trac 6 eye-tracking system was used to record the position of eye gaze throughout the task (see [22] for more specific procedures). The system has maximum accuracy of 0.5 degrees of visual angle, with a resolution of 0.25 degrees. The temporal resolution of the camera was set at 120 Hz. A chin rest was used to reduce head movement. After adjusting the camera of the eyetracker to be centered on the participant’s dominant eye, a short calibration sequence was administered whereby the participant looked at 9 points across the display monitor. After this sequence, the eye-tracking system was able to accurately and continually calculate the participant’s point of gaze relative to the display.

Data reduction.

After the initial 9-point calibration, the accuracy of the system was checked at the beginning of each trial for a period of 1500 milliseconds during which the participant fixated centrally (i.e., fixation cross). An average of the calibration checks across the 40 trials was used to refine the original calibration. Individual trials where the fixation during this 1500 millisecond period were aberrant were excluded. Data from each trial were only analyzed if valid eye data were collected for at least 50% of the duration of that trial. Data from particular subjects were excluded from analysis if that participant lost more than 50% of their trials. Of the 60 participants recruited, a total of 13 were lost due to this system. This system of data reduction eliminates erroneous and missing data due to issues with calibration, mechanical problems, and experimenter error [22]. Our final sample size was 47 participants, with 22 in the Depressed group and 25 in the Anxious group.

Eye fixation variables.

Eye movement parameters reflecting the topographical characteristics of scanning behavior were the proportion of fixations on previously defined regions of interest of the face: entire face, eye region, and mouth region. For each trial, we calculated average percent fixation data in the eye region relative to data in the face, and mouth region relative to face [30]. For this reason, we did not remove eye and mouth regions when examining fixations in the face, as data was calculated based on proportion of fixations relative to the entire face. Fixations were defined as the participant keeping their gaze within a 1 degree area for at least 100 milliseconds. Thus, from the reduced eye data we calculated proportion of fixation data relative to other data (i.e., sum of time length of each fixation during the trial divided by length of the trial) in each region of interest (i.e., face, eyes, mouth) for each face type (i.e., happy, sad, angry, fearful). For example, “Eye/face” is the proportion of fixation data captured in the eye region relative to the proportion of data captured in the face region, and “Mouth/face” is the proportion of fixation data captured in the mouth region relative to proportion of data captured in the face region [30]. Table 1 presents proportion of fixation data relative to other data in each region of interest (i.e., face, eyes, mouth) for each face type (i.e., happy, sad, angry, fearful). We also took the average of all emotional conditions to calculate a grand mean of face, eye, and mouth proportions for the fixation data. Since, for the majority of each trial, the emotional faces are indistinguishable from each other (in fact, the subject ends the trial immediately once they are able to discern what emotion the face bears), this represents an additional primary outcome measure, see Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of visual attention by diagnostic group status.

https://doi.org/10.1371/journal.pone.0250176.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Correlations between fixations and symptom severity scores.

https://doi.org/10.1371/journal.pone.0250176.t002

Data analyses.

RStudio was used to conduct analyses. We conducted a series of regression analyses to test our hypotheses and control for covariates. We used a conservative Bonferroni correction of p <.017 to account for multiple comparisons. For Hypotheses 1 we conducted t-tests to compare eye fixation data between depressed and anxious groups. For Hypothesis 2, we regressed depression severity (BDI-II) onto fixations in the eye region of sad faces. For Hypothesis 3, we regressed anxiety severity (BAI and SIAS) onto proportion of fixations on the entire face, the eye region, and the mouth region of fearful faces. For Hypothesis 4, we regressed anxiety severity onto fixations in regions of interest in happy faces. As mentioned, these data are part of a larger randomized controlled study [22] where half of the participants received oxytocin and half received placebo before completing the emotion recognition task.

We used R pwr package to verify that our sample was adequately powered to detect significance after lost participants due to calibration errors and missing eye data. Our final sample size of 47 provided adequate power (β = .80) to detect effects of medium to large size (f² = .25; α = .05) using multiple regression (Cohen, 1992) and t-tests (d = .58, α = .05).

Results

Descriptive statistics are presented in Tables 1 and 2. With regard to emotion recognition performance, average accuracy scores for all faces in our sample was.93 (SD = .06, range .78 –1.0)[happy = .999, angry = .84, fearful = .93, sad = .97) while intensity scores averaged 40.57 (SD = 9.10) and ranged from 21.03—70.77)[happy = 29.40, angry = 46.25, fearful = 43.61, sad = 44.83]. There were no differences in accuracy or intensity scores between the Depressed and Anxious groups (p >.05). The differences between ER accuracy and intensity scores based on drug status (oxytocin vs. placebo) between Depressed and Anxious groups is the subject of prior work [22]. The average BAI score in our sample was 18.20 (SD = 9.60, range = 0-44), indicating moderate anxiety. The average BDI score in our sample was 19.37 (SD = 10.99, range = 0—41), indicating mild depression. The average SIAS score in our sample was 35.42 (SD = 19.64, range = 0-77), indicating moderate social interaction anxiety.

Before proceeding with any analyses, we tested to see that there were no significant differences in eye gaze patterns between participants who received oxytocin (n = 22) and those who received placebo (n = 25) in our sample using a series of Welch Two Sample-tests. Fig 1 displays the proportion of fixations in the face, eye, and mouth regions for all emotional faces (happy, sad, angry, and fearful), plotted by anxiety severity (BAI) and placebo and oxytocin status. There were no significant differences in proportion of fixations in any region of the face for any emotion between the oxytocin and placebo groups (p range.11 in mouth region of happy faces to.80 in sad faces). We then tested the relationship between a mood disorder and fixations on negative faces, expecting more fixations on sad, angry, and fearful faces if the participant was depressed vs. anxious using a series of Welch Two Sample-tests (Anxious only n = 25; Depressed n = 22). None of these were significant (p range.21 for proportion in fearful faces to.44 for proportion fixations in sad faces).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Proportion of fixations in regions of interest by Beck Anxiety Inventory (BAI) scores and oxytocin/placebo status.

The black circle represents participants who received placebo. The white circle represents participants who received oxytocin. BAI = Beck Anxiety Inventory.

https://doi.org/10.1371/journal.pone.0250176.g001

We expected that higher depression scores based on BDI-II would be associated with more fixations on sad faces. Results showed that higher depression scores were not significantly associated with increased fixations in sad faces overall (p = .7), or in the eye (p = .17) or mouth regions (p = .30) of sad faces. Thus, contrary to our hypothesis, depression severity did not predict fixations on sad faces. To follow up on this unexpected finding, we tested the relationship between depression severity and fixations on angry, happy, and fearful faces, and found that depression severity did not predict any fixations on negative (angry, fearful) or positive (happy) faces (p >.05).

Next, we tested the hypothesis that higher anxiety scores would be correlated with more fixations on threat, i.e., fearful faces. Using fearful faces, we examined the relationship between proportion of fixations on the entire face, the eye region, and the mouth region using regression. Results showed that higher anxiety scores based on BAI were associated with more fixations in the mouth region of fearful faces (R² = .09, F (1, 41) = 4.18, β = .30, p <.05), although the effect size was small (f² = .10), and did not survive Bonferroni correction. The relationship between level of anxiety and proportion of fixations in the entire face and eye region of fearful faces was not significant (p >.05). We then examined the connection between levels of social anxiety, based on the SIAS, and fixations in regions of interest. Results showed that higher levels of social anxiety were associated with decreased attention to the eye region of fearful faces (R² = .12, F (1, 41) = 5.48, β = -.34, p <.05, f² = .14). This did not remain significant after Bonferroni correction. Higher levels of social anxiety did not predict fixations in the mouth or entire face (p >.05), see Table 2.

To test our last hypothesis, we examined the relationship between anxiety severity and eye gaze fixation using multiple regression, with the hypothesis that higher anxiety severity (based on BAI) would be associated with reduced fixations in happy faces. Higher anxiety scores did not predict reduced fixations on happy faces overall, or in the eye region of happy faces. There were, however, significant differences in attention to the mouth region for happiness and other emotional faces, described below.

To examine the relationship between BAI and eye gaze fixations, we controlled for depression status, thinking that the relationship between eye tracking and anxiety may be changed by the presence of mood disorder. Additionally, although we had already established that there was no main effect for oxytocin on eye gaze variables, we controlled for oxytocin here to confirm that there was not an interaction effect of drug and anxiety severity. Results showed that higher anxiety (BAI) was associated with increased fixations in the mouth region of happy faces (R² = .18, F (3, 39) = 2.86, β = .38, p<.05) and mouth region of sad faces (R² = .20, F (3, 39) = 3.28, β = .39, p<.05), with small effects range f² = .22—.25, which survived Bonferroni correction. We also explored the relationship between social anxiety severity and visual attention, and found that increased social interaction anxiety, as measured by the SIAS, was associated with significantly decreased fixations in the eye region of happy faces (R² = .23, F (3, 39) = 3.81, β = -.38, f² = .30), which was significant after Bonferroni correction. Additionally, increased social anxiety severity was associated with decreased fixations in the eye region of fearful faces (R² = .19, F (3, 39) = 3.11, β = -.31, p<.05), f² = .23), which did not remain significant after accounting for multiple comparisons using Bonferroni methods. The relationship between social anxiety severity and attention to the eye region of sad faces approached significance (R² = .18, F (3, 39) = 2.82, β = -.29, p = .051). When oxytocin was removed as a covariate, this finding was significant: social anxiety severity predicted decreased attention to the eye region of sad faces (R² = .14, F (2, 40) = 3.39, β = -.29, p<.05). Interestingly, this indicates that oxytocin may have a particular effect on visual attention in individuals with higher levels of social anxiety, even though oxytocin failed to produce a main effect on visual attention overall. This finding is consistent with the research using oxytocin to enhance treatment of social anxiety disorder [31]

Discussion

While a large body of literature has shown that emotional disorders are characterized by attentional biases for emotional stimuli, to our knowledge, this study was the first to examine the impact of symptom severity on attention to specific regions of interrest in emotional faces in a comorbid and transdiagnostic entirely clinical sample. This advance from the existing research is notable, given high rates of comorbidity between anxiety and mood disorders [32] and lack of consensus in the field as to whether biases in attention represent a general or specific deficit in emotion recognition [33, 34].

Comorbidity

Our results showed that having a depression diagnosis vs. anxiety only diagnosis did not significantly impact visual attention: there were no differences between attention to sad, happy, angry, or fearful faces based on depression status. Next, results showed that higher anxiety scores based on the BAI were associated with more fixations in the mouth region of happy faces and sad faces, but not angry or fearful faces. In contrast, higher SIAS scores did not predict visual attention to the mouth region. Higher SIAS scores were associated with decreased fixations in the eye region of happy faces.

Acknowledgments

The authors thank Bonnie Brown, Alice Cronin-Golomb, and Kristin Long for their assistance with this study.