Introduction
With more than 300 million people suffering from depression worldwide, it is amongst the most prevalent mental disorders and, according to the World Health Organization (WHO
2017), has become the leading cause of disabilities, with a major contribution to the overall global disease burden. Despite the variety of existing treatment options for depression, relapse rates are high (e.g., about 30% as reported in Seemüller et al.
2014), with an increasing risk of recurrent depression with each subsequent depressive episode (Steinert et al.
2014). This suggests that the underlying mechanisms which maintain depression are not very well understood yet, and hence may not be targeted sufficiently by current treatment programs.
A potential cognitive vulnerability factor which is thought to maintain depressive symptoms and predispose individuals to repeatedly develop new episodes, is the heightened attention to depression-relevant, negative information compared to positive information, commonly found in depressed individuals (e.g., Armstrong and Olatunji
2012). Research making use of eye-tracking technology suggests that in depression, this so-called negative attentional bias is specifically characterized by difficulties with disengaging attention from negative stimuli once they have become the focus of attention (Sanchez et al.
2013). At the same time, depressed individuals show reduced maintained attention to positive stimuli, compared to healthy individuals (e.g., Ellis et al.
2010; Kellough et al.
2008; Sears et al.
2010).
According to cognitive theories of depression, attentional biases play a causal role in both development and maintenance of this disorder (Beck
1976; Teasdale
1988). As a consequence, computerized training paradigms have been developed, aiming at reducing depressive symptoms through altering maladaptive information processing tendencies; the so-called cognitive bias modification techniques (CBM; Mathews and MacLeod
2005). The most frequently used paradigm for modifying (and measuring) attentional processes is the dot-probe task (MacLeod et al.
2002). On each trial of this task, two stimuli are presented simultaneously on the computer screen, usually a negative picture and a positive (or neutral) picture (or word). After a short delay, both stimuli disappear and a target replaces one of the two stimuli, which participants have to react to. Faster reactions to targets replacing negative compared to positive stimuli indicate a negative attentional bias. In the training version of this task, the targets replace mostly the positive (or neutral) stimuli, such that participants’ attention is selectively trained away from the negative stimuli.
To date, the dot-probe task has mainly been used to assess and modify attentional bias in anxious populations (for review see Cisler and Koster
2010). Only a few studies have tried to modify attentional processes in depressed individuals, with inconsistent findings so far. While some studies managed to reduce the bias toward negative stimuli and accordingly the depressive symptoms, many other studies failed to modify an attentional bias and to replicate the beneficial therapeutic effects on depressed mood or symptoms, questioning the efficacy of attentional bias modification (ABM) procedures for depression (for reviews see Cristea et al.
2015; Mogoaşe et al.
2014). In a paper reflecting on the increasing number of ABM failures Clarke et al. (
2014), however, cautioned against taking the absence of evidence in some studies as evidence against the theoretical basis of ABM in general. The authors drew attention to the fact that most studies that succeeded in modifying an attentional bias also induced emotional change. By contrast, those studies that failed to modify an attentional bias also failed to find beneficial effects on mood. This suggests that conventional ABM paradigms may not be optimal for reliably modifying and measuring attentional bias, and that more research is needed into the task conditions under which ABM actually changes attentional processes. According to this argument, more promising clinical applications of ABM depend on the development of more effective attention modification procedures.
One of the most frequently mentioned limitations of the dot-probe and other reaction time (RT) based ABM paradigms is the low reliability of these tasks (e.g., Schmukle
2005; Waechter and Stolz
2015). More importantly, however, the suitability of the dot-probe task has especially been doubted in the context of depression, as it is not clear which component of attention is measured and trained (Leyman et al.
2007). It has been argued that with the longer stimulus durations commonly used in depressed populations (Beevers et al.
2015; Wells and Beevers
2010), participants may shift their attention back and forth between stimuli before the target appears. This in turn leaves undetected whether the task is tapping into heightened vigilance for negative stimuli or the depression-characteristic slowed disengagement from negative stimuli (for a more thorough discussion of the limitations of the dot-probe task, see Ferrari et al.
2016).
Based on the limitations of the dot-probe task and other RT-based ABM paradigms, Ferrari et al. (
2016) developed a new ABM paradigm, incorporating eye-tracking technology. This eye-tracking based attentional bias modification (ET-ABM) paradigm allows for the continuous assessment of eye-movements, and hence for a potentially more reliable assessment and training of the specific attentional components that are biased in depression. During this ET-ABM task, participants are trained to disengage their attention from negative pictures and to keep their attention on positive pictures. On each trial of the task, participants first have to fixate a cross on a computer screen, after which two positively and two negatively valenced pictures are presented in a 2 × 2 grid. Importantly, the trial only continues after a sufficiently long fixation of a positive picture (1000 ms). Hence, if a negative picture replaces the cross, the trial only continues when participants look away from the negative picture and fixate one of the positive pictures. If a positive picture replaces the cross, the pictures disappear when participants keep their attention on the fixated positive picture or fixate the other positive picture. As soon as a positive picture has been fixated for 1000 ms, all pictures disappear and a target stimulus is presented at the location of the last fixated picture. The target has to be identified by pressing a corresponding button. Different from previous RT-based ABM paradigms, the training trials in this task only continue if participants show the required viewing pattern, tailoring the pace of the task to the individuals’ performance.
In a first proof-of principle study (Ferrari et al.
2016) with an unselected student sample, this positive training was compared to a negative training with the opposite training contingencies. Thus, participants were trained to direct their gaze to negatively valenced pictures. Results of this first study showed that this ET-ABM training is suited to alter attentional processes relevant in depression: The positive training induced a positive sustained attention bias, that is, longer sustained attention to positive than to negative pictures. More specifically, it trained participants to more quickly disengage their attention from negative stimuli and direct their attention to positive pictures. No such changes were found in the negative training group. Although the training affected mood directly, with the negative group showing a stronger increase in negative mood in response to the training than the positive group, the training did not differentially affect emotional reactions to a subsequent laboratory stressor.
Although this first study provides promising evidence that depression-relevant attentional processes can be altered by means of this novel ABM paradigm, several questions remain to be answered. First, in the previous study, the positive training was compared to a negative training in order to maximize group differences, hence it remains to be established whether the positive training is also superior to a placebo control condition in changing the attentional processes. Second, before applying this training in clinically depressed populations, it should first be tested whether the results can be replicated in a sample with subclinical levels of depression, which is assumed to show a stronger pre-existing negative attentional bias. Finally, training effects in the first study were only tested with the same set of stimuli as used during the training, leaving it open whether training effects are restricted to the specific stimuli being used, or whether the attentional processing of positive and negative stimuli in general is affected.
Hence, the primary aim of this study was to replicate the findings of the previous study in an emotionally vulnerable sample, with a sham training as control condition and different stimulus sets in training and assessment. To this end, we randomly assigned participants with elevated depression scores to one of two training conditions. Half of the participants received the positive training (PT) in which they were trained to direct their gaze away from negative pictures and towards positive pictures. The other half received a sham training (ST), where no valence-specific gaze patterns were reinforced. Importantly, at the beginning of the experiment, a negative mood was induced by means of a sad movie. This allows for re-activation of otherwise latent depressogenic structures in emotionally vulnerable individuals (e.g., Beck
1967) and hence may serve to elicit a negative attentional bias in our sample (for a detailed description of the procedure, see Scher et al.
2005). We expected that, compared to participants in the ST, (1) participants in the PT would show an increase in positive attentional bias (i.e., relatively longer fixations on positive than on negative pictures), and that (2) participants in the PT would specifically learn to faster disengage their attention from negative pictures.
To get a first indication of the potential therapeutic effects of the training, we additionally explored participants’ mood reactivity and recovery in response to a laboratory stressor at the end of the experiment. In their eye-tracking experiment, Sanchez et al. (
2013) showed that specifically the slowed disengagement from negative information is related to impaired mood recovery after stress in depressed individuals. In accordance with these findings, we expected that participants in the PT would show better stress recovery than participants in the ST.
Discussion
To address the limitations of previously used RT-based ABM paradigms, a novel ABM paradigm based on eye-tracking was recently developed (i.e., the ET-ABM; Ferrari et al.
2016). This paradigm was specifically designed to assess and target the attentional components that are biased in depression: the disengagement from negative stimuli and the maintained attention to positive stimuli. A first proof-of principle study with healthy students showed that, compared to a negative training version, the ET-ABM can induce a positive sustained attention bias as well as faster disengagement from negative stimuli. The aim of the present study was to replicate these promising findings in an emotionally vulnerable sample of dysphoric students, with a placebo sham-training as control condition.
In line with the findings of Ferrari et al. (
2016), the PT induced a positive sustained attention bias (i.e., longer fixations on positive than on negative stimuli). Notably, both PT and ST showed an increase in positive sustained attention bias. However, this increase was stronger in the PT group, supporting the effectiveness of the training in modifying attentional processes in dysphoric individuals. As in the previous study, these general training effects were again driven by reduced disengagement latencies from negative stimuli in the PT group. Beyond that, the PT group also showed an increase in maintained attention to the first fixated positive pictures, suggesting that the training may also affect the initial processing of positive stimuli. In contrast, the ST did not induce any valence-specific attentional viewing patterns. Instead, participants in this group became generally slower with directing their gaze away from the initially fixated pictures, resulting in slower disengagement from both negative and positive pictures. Summarizing, while the PT was effective in reversing an initially negative attentional bias into a positive attentional bias, characterized by relatively longer fixations on positive pictures and by relatively quicker disengagement from negative pictures, the ST resulted in a decline of the negative bias. A potential explanation for the observed effect in the ST group might be that the ST was possibly more difficult than the PT. Remarkably, more participants in the PT became aware of the reinforced training pattern, and the PT group experienced a stronger feeling of control over the task. It is likely that the ST therefore was more tiring than the PT, resulting in slower latencies in general.
In a previous study (Ferrari et al.
2016), the PT did not modify initial maintained attention to positive pictures. It had been suggested that the temporal criteria defining a fixation as sufficiently long to continue a training trial (i.e., 1000 ms) might not be appropriate to induce “longer” maintained attention to positive stimuli. As our main goal was to replicate the earlier training effects on general sustained attention and attentional disengagement from negative stimuli, we did not increase the required fixation duration on positive pictures. Nevertheless, in the current study, training effects were partially driven by the increased initial maintained attention to positive stimuli. This might be explained by the different stimulus sets used in the two studies. In the previous study, negative and positive pictures were matched on their valence ratings, whereas this was not done in the current study. As a result, more extreme positive and negative pictures might have been presented during the training, which possibly resulted in a better contrast between these valences. This change in contrast might have helped participants to more easily identify the two different responses required to react to the two different picture types, resulting in the modification of both indices, disengagement from negative pictures and maintained attention to positive pictures. Although this explanation remains speculative, the current findings provide promising evidence that the ET-ABM may actually directly tap into both components of attention that are relevant in depression. Importantly, as different picture sets were used during training and assessments, we may further conclude that the observed training effects are not merely the result of stimulus-specific response patterns, but reflect a modified attentional processing of emotionally valenced information in general.
To get a first indication of the potential therapeutic effects of the ET-ABM, we additionally explored changes in mood. In contrast to the first study, the PT and ST did not differentially affect participants’ mood state. In general, positive mood increased indistinguishably in both groups from before to after the training, possibly pointing to recovery from the negative mood induction at the beginning of the experiment. More importantly, the training did not affect mood changes during the stress task either. In line with the previous study (Ferrari et al.
2016), PT and ST groups did not differ in their mood reactivity or recovery from the speech challenge. It was therefore speculated that stress-attenuating effects of the training may be restricted to emotionally vulnerable samples, as suggested by previous CBM research (Becker et al.
2016). Our sample did consist of individuals with elevated depression scores. Hence, one possible interpretation of our results may be that in depression, the modification of attentional processes does not affect mood reactivity or recovery.
In the context of anxiety, the link between attentional bias and emotional vulnerability has been investigated in a range of studies (for a review, see Clarke et al.
2014), whereas only a few studies have addressed this topic in depression. Although Sanchez et al. (
2013) did not experimentally manipulate attentional processes, they found that slower disengagement from negative stimuli predicted lower mood recovery after stress. Together with the few studies showing that reducing a negative attentional bias may attenuate depressive symptoms (Browning et al.
2012; Wells and Beevers
2010; Yang et al.
2014), this provides evidence supporting a causal link between attentional bias and maintenance of depressed mood. It is important to note here, that the latter studies all made use of multiple training sessions, distributed over a longer period of time. However, given the limited number of studies and the contradicting conclusions from meta-analyses regarding number of sessions as a moderator of training effects (Beard et al.
2012; Cristea et al.
2015; Hallion and Ruscio
2011; Mogoaşe et al.
2014), it remains to be investigated whether training effects on stress reactivity and recovery, or even depressive symptoms, can be achieved by increasing the number of training sessions.
Moreover, we would like to emphasize that the stress task employed in this study may not be optimal for measuring transfer effects of the training to mood responses, which might be the reason why we found no relation between attentional processing and emotional reactivity. Although Sanchez et al. (
2013) found a significant association of slow disengagement from negative stimuli with lower mood recovery after a speech challenge as used in our study, one may question whether a performance-related speech-challenge can be considered a relevant stressor in the context of depression. In fact, the link with attentional processes was exclusively found for “sad” mood recovery. Hence, before drawing firm conclusions about the causal role of attentional bias in depressed mood, future research should consider to increase the number of training sessions and investigate its effect on mood recovery after a depression-relevant stressor. Such a stressor could for instance involve a video-clip that induces sad mood, as used at the beginning of our experiment.
Subsequent studies using this paradigm might also want to include other measurement instruments that can detect far transfer effects. Even though this study administered a free-viewing assessment task with different pictures than used during training, the free-viewing task shares several characteristics with the training paradigm. To rule out that only a single task-relevant component has been trained, we recommend to make additional use of alternative bias measures, such as a spatial-cueing task (Baert et al.
2010) or the engagement-disengagement eye-tracking task by Sanchez et al. (
2013). Moreover, as the current study does not allow to attribute training effects to either of the two attentional components (i.e., negative disengagement or positive maintained attention), follow-up studies are required to disentangle the specific working mechanism of the ET-ABM. Finally, a measurement-only control condition might be a useful addition for future research, as sham-training procedures as implemented in this study may have training unspecific effects as well (e.g., Gladwin
2017; Wells and Beevers
2010).
For these follow-up studies, we would strongly recommend to take all three attentional components into account (i.e., sustained attention, negative disengagement, positive maintained attention). The current study found no sustained attention bias at baseline, which might have been expected based on the extensive literature on the existence of negative attentional biases in depression (e.g., Peckham et al.
2010). However, the fact that we found a slower disengagement from negative than from positive pictures points to the importance of looking into the different attentional components separately, rather than exclusively investigating a general bias. Differentiating between negative disengagement and positive maintained attention may allow for a more sensitive measurement of attentional bias in depression.
To summarize, this study provides further support for the effectiveness of the ET-ABM in modifying the specific attentional components assumed to be causally involved in the development and maintenance of depression. In order to investigate whether ABM can indeed alleviate emotional vulnerability, it has been suggested that we need refined or new paradigms which reliably assess and modify these processes (Clarke et al.
2014). This study suggests that the ET-ABM task may indeed be such a paradigm that facilitates further progress in this field of research. Fortunately, the technical developments of the past decade made eye-tracking devices accessible at reasonable prices and suitable for the use in hospitals without requiring the expertise of technicians. Therefore, the next step should be to investigate the beneficial effects of the ET-ABM on clinically relevant measures in a patient sample.