The key search terms used were “depression,” “depressive disorder,” “MDD,” “eye tracking,” “eye movement,” and “attention bias (AB).” A list of synonyms was created for each key term. The key terms were searched along with their synonyms in each separate database. The databases searched included PubMed, Web of Science, ScienceDirect, EBSCOhost, The Cochrane Library, Embase, China National Knowledge Infrastructure, Wanfang Data, and VIP database. The search window extended from when the databases were created to March 2022.

Two of the authors (Huang and Zhu) conducted independent systematic literature and manual searches of the reference list of the previous meta-analysis, and the retrieved articles were assessed.

The inclusion criteria of this meta-analysis were as follows: (a) studies that published case–control analysis on eye tracking of individuals with depression, (b) patients aged ≥ 18 years who met the current international diagnostic criteria (Diagnostic and Statistical Manual of Mental Disorder 5 [DSM-5] or International Classification of Diseases) or were clinically diagnosed with depression as the case group and healthy individuals as the control group, (c) studies that used free-viewing tasks, and (d) studies that contained statistics and sufficient data for analyses.

Exclusion criteria were (a) review articles or meta-analyses, case reports, (b) eye movement studies examining patients with other disorders, and (c) articles that did not meet the diagnostic criteria for patients with depression in the DSM, ICD, or depression Diagnostic Scale, questionnaire (i.e., BDI, HAMD, PHQ-9).

After removing duplicates, the titles and abstracts of the screened studies were independently screened by authors (Huang and Zhu) and any non-pertinent studies were excluded. The final list comprised studies accepted by both the authors. If the inclusion criteria were met, the full-text article was retrieved and screened to evaluate the available data for analysis.

Data extraction was independently performed by two authors (Huang and Zhu). Disagreements were resolved via discussion with another author (Li) until consensus was reached.

Data extraction included the following: (a) basic information regarding the included literature: title, first author, and publication time; (b) basic characteristics of the patients: diagnostic criteria for depression, average age, number of samples in each group, and stimulation materials; (c) eye tracking parameters applied in the free-viewing tasks: fixation duration, dwell time (in ms or %), and number of fixation time (in frequency or %).

Review Manager version 5.4.1. was used for statistical analysis and generation of forest and funnel plots. The effect value was expressed via standardized mean difference (SMD) along with 95% CI; larger the absolute value of SMD, larger the effect size. Chi-square test was used for heterogeneity test, and I² ≤ 50% and P >0.10 were considered low heterogeneity. According to the research theory and purpose, this study adopted the random effect model (Borenstein et al., 2009).

Q statistic was used to determine if subgroup analysis was necessary, as significant heterogeneity suggested that there were factors accounting for between-study variance in the effect size. A priori moderator variables included patient characteristics (age) and experimental characteristics (stimulus duration and emotional stimulus type).

To evaluate for publication bias and the stability of the results, we conducted sensitivity analysis and constructed separate funnel plots to determine fixation duration, number of fixation, and first fixation duration (Jin et al., 2015). We assessed the risk of bias using NOS Literature Quality Evaluation Scale for all the included studies (Stang, 2010).

The search strategy identified 13,198 articles, which was reduced to 9,214 after excluding duplicates. After screening the titles and abstracts of these articles, 96 were included for full-text screening and 82 were excluded. In total, 14 articles met the inclusion criteria and were included in the meta-analysis (see Figure 1).

These 14 eye tracking studies included 474 patients with depression and 693 healthy controls; participants were arranged in terms of age, and only two studies included older individuals. All the studies involved free-viewing tasks. In these studies, the participants were asked to simply observe emotional stimuli while they were shown faces or images. The emotional stimuli included being happy, positive, neutral, negative, sad, dysphoric, threatened, or angry. The duration of stimulus presentation ranged from 4 to 30 s. The outcome measures included early and later attentional processes (see Table 1).

Publication bias was measured using funnel plots, and potential heterogeneity was eliminated using the one-by-one exclusion method. Publication bias minimally affected the results, and sensitivity analysis established the conclusion as relatively robust (see Figures 2–5).

The quality of the included studies was determined using the 10-point Newcastle–Ottawa scale (NOS; Table 2). Of the analyzed studies, two scored 5 points, six scored 6 points, and six scored 7 points. Thus, the high quality of the literature indicates that publication bias exhibited minimal effect on the results.

Fixation duration has been considered indicative of the depth of individual cognitive processing and maintenance of a stimulus. Our meta-analysis identified that the effect sizes of fixation time for negative and positive emotional stimuli reached 0.59 and −0.71, respectively. This suggests moderate differences between patients with depression and healthy individuals in terms of attention maintenance for negative and positive emotional stimuli. Our results are similar to those reported by Suslow et al. (2020). A study regarding the “antireward center” mechanism of depression reported that the abnormal increase in the activity of the lateral bridle nucleus neurons in patients with depression does not only prevent the processing of positive emotional information but also inhibits the positive processing of reward stimuli. Furthermore, the insensitivity of such patients toward reward stimuli impacts the goal-directed attentional system (Gold and Kadriu, 2019). This confirms that patients with depression exhibit insufficient bias toward positive emotional information and excessive cognitive processing and maintenance of negative emotional information.

The number of fixations primarily reflects the information extraction efficiency and information processing ability of the cognitive process. Patients with depression looked at fewer positive images compared with healthy individuals. The effect sizes of the fixation times for negative and positive emotional stimuli (SMD = 0.32 and −0.87, respectively) suggest moderate differences between patients with depression and healthy individuals with respect to attention maintenance in response to negative and positive emotions. The results of event-related potential studies (Ke et al., 2020) also confirmed that patients with depression exhibited electroencephalogram characteristics of prolonged response time when processing negative emotional information. This indicates that patients with depression extract negative emotional information with low efficiency and poor information processing ability.

The first fixation duration chiefly reflects the early attention given to directional processing characteristics (Grafton et al., 2016). No significant group differences between patients with depression and healthy individuals were observed for initial attention orientation to negative, neutral, or positive emotional stimuli, consistent with previous findings (Grafton et al., 2016). Thus, patients with depression appear to not be characterized by alterations during early orientation toward emotional stimuli (Suslow et al., 2020).

Difficulties in separating negative emotional information may be an important reason underlying the onset and maintenance of depression. In addition, functional brain imaging studies have demonstrated that the intensity of cortical activity, such as in the right prefrontal and right parietal lobe, weakens in depression, and that this cortical function is primarily associated with functions such as attentional selection and transfer (Skinner et al., 2018; Yaroslavsky et al., 2019). However, the electrophysiological indicators involved in the two-stage model framework of emotional attention regulation suggest that patients with depression mostly exhibit an enhanced P3 amplitude and negative bias during early attention allocation (Ao et al., 2020). However, controversy remains concerning the occurrence stage of negative attention bias, and this may be related to the lack of consistency in the emotional stimulus types and presentation time in the various studies. Hence, more evidence regarding the onset stage of negative attention bias is required.

The subgroup analysis revealed a significant effect of age. Patients with depression who were young and middle-aged observed more negative emotional stimuli (SMD = 1.21) and less positive emotional stimuli (SMD = −0.85) than healthy individuals, although no significant difference between older healthy individuals and older patients with depression was observed. According to the socioemotional selectivity theory (Carstensen et al., 1999; Carstensen, 2006), older adults prioritize emotion-regulatory goals. Furthermore, individual attentional sensitivity to negative stimuli gradually decreases with age (Mather and Carstensen, 2003; Lu et al., 2017); structural and functional changes in the brain also follow the same trajectory of behavioral change (Luna and Sweeney, 2004). These findings confirm the presence of emotional attentional biases in patients with MDD that are influenced by age. Attention bias modification (ABM) techniques have been confirmed to effectively improve cognitive bias modification procedures in depression (Vazquez et al., 2016). Consequently, clinicians should select the corresponding intervention based on patient age to improve clinical efficacy of ABM techniques. However, as there are presently relatively few empirical studies regarding eye movements in older patients with depression, more studies are warranted to confirm these preliminary results.

The subgroup analysis revealed a significant effect of negative emotional type. For instance, the effect size of fixation time on the emotional stimuli such as sadness and anger was 0.86 and 0.98, respectively, indicating a more significant attentional bias toward these stimuli. Daily negative emotions persist in depression, especially constant sadness (American Psychiatric Association, 2013). In addition, patients with MDD exhibited inflexibility in sadness and avoidance, which were associated with weaker FC between the right sgACC and pregenual/dorsal anterior cingulate (pg/dACC) (Schwartz et al., 2019). Furthermore, the emotions sadness and anger have been highly associated with depression (Koh et al., 2002; Ou and Hall, 2018). According to clinical surveys, the irritability rate of patients with depression can be as high as 30–85% (Bodner et al., 2018), and the DSM-5 views anger as a marker of depression in children and adolescents (American Psychiatric Association, 2013). Anger and irritability are prevalent in adults with depression and hinder the correction of depression cognitive bias more strongly than other negative emotions (Perlis et al., 2009). Our results are similar to those of previous studies, thus verifying that sadness and anger play an important etiological role in developing depression, which further helps simulate the treatment of emotion regulation and attentional bias correction in depression.

The interaction effect of the stimulus presentation time on emotional stimuli was not significant. Our results are similar to those of a meta-analysis conducted by Peckham who found no evidence regarding the significant moderation of biased attention toward negative information by stimulus presentation time in emotional Stroop or dot probe tasks (Peckham et al., 2013). Total gain duration and early stage of automatic attention allocation are both indicators of attention bias. Total gain duration usually used as attention maintenance index to evaluate negative attention bias (Duque and Vázquez, 2015; Lazarov et al., 2018), and it is reported that there was a significant main effect of stimulus presentation time on total gain duration (Wells et al., 2014). Previous ERP studies have demonstrated that automatic attention attraction by emotionally stimuli could reliably enhance the P1 amplitudes in the dot-probe task, which has been associated with rapid and unconscious bottom-up attention allocation (Luck and Kappenman, 2012), representing a facilitated perception in response to the emotional stimulus (Brosch et al., 2008; Liu et al., 2015). Eye tracking study in patients with depression have found a bias in the initial orientation to depression-related images (Sears et al., 2011). There is a lack of consensus on whether negative attentional bias occurs at early attention stages. It is demonstrated that patients with depression show a attentional bias toward congruent sadness at 100 ms, indicating that negative attention bias occurs in the early stage of attention (Trapp et al., 2018; Ao et al., 2020). While Cao found that the negative attention bias of patients with depression appeared after 6 s, suggesting that attention bias does not occur until the later stage of cognitive processing (Shuang et al., 2019). Therefore, we suggest that stimulus presentation time may be an influencing factor of emotional stimulus processing in patients with depression. But our study results are inconsistent with these results, probably owing to the stimulus material or material presentation time grouping. Therefore, these factors should be controlled and further discussed in the future.

In addition, attention bias mechanism has been suggested to mostly occur when processing social information stimuli. Faces showing emotions reflect the emotional state of others and thus contain more social information than emotional scene pictures and are more likely to attract the attention of patients with depression (Gilboa-Schechtman et al., 2004). However, there was no significant difference in the attentional bias toward emotional faces or scenes among the patients with depression included in this study. According to Leyman's research on attentional bias using forward-placed and inverted emotional faces, it is the emotional titer expressed by stimulus materials that causes individual attentional bias (Leyman et al., 2011) and when we looked back at the emotional scenes that were included in the study, we found that some of the emotional scene materials contained both characters and scenes, which is the same emotional valence as the emotional faces. Our results may support the idea that it is the emotional valence expressed by the stimulus material that causes attention bias in patients with depression. We suggest that future studies can compare the attention bias of depressed patients to emotional faces and emotional scenes to verify this view.

Eye movements are sensorimotor functions of the brain, and the effect of gender on attention patterns has been shown in autism spectrum disorders (Harrop et al., 2018). Classification of emotions based on EEG and eye movement signals also agree that emotion classification differs between male and female, and there are larger differences of brain activities between male and female for sad, which is closely associated with depression (Bao et al., 2019). Studies on attention bias in healthy adults also show that attention bias is influenced by gender and mainly manifests in the regulation function of negative stimulus avoidance (Pintzinger et al., 2016). Previous studies on attentional bias in depression have shown that sex differences in the subjects may lead to different initial attentional bias, and female patients with depression are more likely to have an initial attentional bias toward negative information (Sears et al., 2011; Li et al., 2016) and compared with men, women also have longer depressive episodes and are also presented higher levels of inflammatory, serotonergic markers, and neurotrophic (Oquendo et al., 2013; Labaka et al., 2018). Thus, it is important to consider gender as a biological variable in clinical depression studies, which may facilitate the development of more effective diagnostic methods and pharmacotherapies for depression.

Furthermore, the results of subgroup analysis of diagnostic methods showed that there was no significant difference in the negative attention bias of depressive patients diagnosed based on clinical diagnostic criteria and questionnaire, and we preliminarily believed that the diagnostic methods did not have much influence on the results. However, in order to ensure the reliability and extensibility of the research results, the inclusion criteria should be more strictly established in future studies.

Several limitations must be considered before translating these results into clinical settings. First, we only analyzed the attention-processing characteristics of emotional stimuli in patients during the onset of depression. Thus, our study did not summarize these characteristics in the different stages of depression, such as first recurrence and disease duration, and did not comprehensively evaluate the value of this parameter for diagnosing depression. Second, only a limited number of studies was included in the meta-analysis for older patients with depression, thereby potentially reducing the reliability of its conclusions. To stimulate the clinical application of attention bias correction technology, future studies should observe and compare the attention-processing characteristics of patients with depression on emotional stimuli at different ages and stages and degrees of the disease, explore more indicators, such as saccade, and further evaluate the application value of eye movement indicators in diagnosing depression.