The Attentional Bias Modification Approach to Anxiety Intervention

Impact of single session ABM in participants without dysfunctional anxiety

Some single sessions ABM studies have examined the impact of attend-threat and avoid-threat ABM on participants’ responses to a worry manipulation carried out in the laboratory. Krebs, Hirsch, and Mathews (2010) and Hirsch et al. (2011) each carried out experiments of this nature. In both studies, attentional assessment revealed that the ABM training manipulation successfully induced a group difference in attentional selectivity, reflecting reduced attention to threat in those individuals who received the avoid-threat ABM condition compared with those who received the attend-threat ABM condition. The successful induction of this intended group difference in attentional selectivity elicited a corresponding group difference in the degree to which a subsequent period of instructed worry served to elicit negative thought intrusions. Specifically, those participants who had received the avoid-threat ABM condition, relative to those who had received the attend-threat ABM condition, showed reduced evidence of worry-elicited negative thought intrusions. These findings suggest that selective attentional responding to threat causally contributes to variation in the degree to which people experience worry-related negative thinking.

Findings from a single session ABM study carried out on younger participants, midrange on a measure of trait anxiety, indicate that variation in this attentional bias may contribute to variation in dispositional anxiety within children also. Using a probe task variant that employed emotional face pairs as stimuli, Eldar, Ricon, and Bar-Haim (2008) delivered intended ABM training to a sample of such children aged 7 to 12 years. Half the participants received the task in the avoid-threat condition, whereas half received it in the attend-threat condition. Pre and post measures of attentional bias, taken using the assessment version of the attentional probe task, verified the success of this intended ABM manipulation in differentially modifying attentional selectivity as intended. Following this successful attentional manipulation, all children were exposed to a stressor that required completion of a difficult puzzle. Anxiety reactivity to this stressor, as assessed by both self-report and behavioral measures, differed significantly as a function of ABM training condition. Specifically, although children who had received the attend-threat ABM condition reported a significant elevation of anxiety in response to this final puzzle stressor, no such stressor-elicited anxiety elevation was reported by those who had received the avoid-threat ABM condition. These latter children also exhibited less behavioral signs of anxiety than the former children while completing the puzzle task.

Two single session ABM studies of this type have failed to achieve the intended modification of attentional selectivity, both using similar probe task variants employing expressive faces as stimuli. One of these studies was conducted on participants who scored midrange on a measure of social anxiety (Klumpp & Amir, 2010), whereas the other used unselected students as participants (Arditte & Joormann, 2014). In both studies, half the participants were given the intended ABM training procedure in the avoid-threat condition, and half were given it in the attend-threat condition. Attentional bias was measured immediately after this intended training procedure, using the conventional attentional probe assessment task in Klumpp and Amir’s experiment, and an eye movement assessment task in Arditte and Joormann’s study. In neither experiment was there any evidence of any training-induced difference between the two training groups in terms of their selective attentional responding to threat. Furthermore, in both studies this failure to modify attentional bias was accompanied by the absence of any training group difference in anxiety reactivity to a final speech task stressor. This shows that when an intended attentional training procedure does not serve to successfully modify attentional bias to threat, then neither does it influence anxious disposition.

This general pattern of findings, showing that successful modification of attentional bias influences anxiety reactivity to a stressor whereas intended training procedures that fail to achieve this attentional objective do not affect such anxiety reactivity, supports the hypothesis that attentional bias causally contributes to anxiety vulnerability. However, it appears that tendency to experience spider fear may be an exception. Spider phobia is characterized by an attentional bias toward spider-related information (Watts, McKenna, Sharrock, & Trezise, 1986). Nevertheless, when Van Bockstaele et al. (2011) carried out an ABM training manipulation that successfully induced differential attention to spider stimuli in unselected student participants, this did not produce a corresponding group difference in spider fear. Specifically, these investigators gave participants a probe task that presented image pairs, each comprising one image of a spider and one of a household object. This task was configured either in the avoid-threat ABM condition or in the attend-threat ABM condition, and a conventional assessment version of the probe task showed that participants receiving these alternative ABM conditions came to differ in such attentional selectivity in the intended manner. Despite this successful modification of selective attention, these groups did not differ in their fear responses when subsequently exposed to a live spider. This suggests that the tendency to display attentional bias toward spider information may be a consequence of elevated spider fear, rather than causal contributor to spider fear. Although Van Bockstaele et al.’s results represent a failure to elicit emotional change following successful modification of selective attention using ABM, these findings nonetheless demonstrate the capacity of the ABM approach to advance theoretical understanding, by revealing which types of anxiety symptoms result in part from the causal influence of biased attentional selectivity, and which do not.

Impact of single session ABM training in participants with dysfunctional anxiety symptoms

These single sessions studies have sought to determine whether the reduction of selective attention to threat can alleviate dysfunctional anxiety symptoms already exhibited by participants. Thus, for example, Hayes, Hirsch, and Mathews (2010) recruited participants based on their self-reported high levels of dispositional worry. These individuals were given a single session of probe ABM training, using word pairs as stimuli, either in the avoid-threat condition or in the standard control condition, which contained no training contingency. Assessment of attentional bias following training confirmed that, relative to participants who received the control condition, those given the avoid-threat ABM training came to show increased attentional avoidance of threat, as intended. Hayes et al. then assessed the impact of this training-induced difference in attentional selectivity on worry-elicited negative thinking, by recording the degree to which a subsequent period of instructed worry served to elicit subsequent negative thought intrusions. For participants in the control condition, the period of instructed worry elicited negative thought intrusions as expected. However, for participants who had been exposed to the ABM training that successfully reduced attention to threat, the period of instructed worry did not elicit negative thought intrusions. These finding demonstrate that, when intended ABM training manipulation successfully reduces attention to threat, then this can attenuate the negative impact of worry in people who report problematic levels of worry.

Similarly positive findings have been obtained using avoid-threat ABM to attenuate dysfunctional symptoms associated with subclinical manifestations of obsessive-compulsive disorder (OCD). Najmi and Amir (2010) recruited students who exhibited high levels of contamination fear and delivered a single session of probe ABM training using word pairs that contained one contamination-relevant (threat) word and one contamination-irrelevant (neutral) word. Half the participants received this task in the avoid-threat condition, whereas half received it in the standard control condition. Attentional assessment revealed that these two groups of participants did not differ in terms of attention to contamination-related information before receiving this procedure, but only those who were given the avoid-threat ABM subsequently developed attentional avoidance of contamination-related information, as intended. Najmi and Amir then had participants perform a behavioral approach task that tested their ability to approach feared contaminant stimuli. Participants who had received the avoid-threat ABM training, that had proven successful in reducing attention to contamination-relevant information, were able to approach these feared contaminants more closely than participants who had received the control condition. Thus, the reduction of attention to threat served to attenuate the severity of this dysfunctional anxiety problem too, when symptom severity was assessed within the laboratory setting.

A single session of avoid-threat ABM also has been shown to reduce problematic social anxiety symptoms in the laboratory. Amir, Weber, Beard, Bomyea, and Taylor (2008) recruited participants who reported problems with public speaking. These participants completed a single session of a probe task that employed pairs of faces as stimuli, one displaying the disgust expression commonly characteristic of social contempt (threat), and the other showing a neutral expression. One group of participants completed this task in the avoid-threat ABM condition, whereas another group completed it in the standard control condition. An attentional assessment procedure, using word stimuli, confirmed that the participants who received the avoid-threat ABM condition came to display significantly reduced attention to threatening social information, relative to those in the control condition. All participants were next exposed to a social stressor that involved giving an impromptu speech. Compared with participants in the control condition, those who had completed the avoid-threat ABM training reported experiencing lower levels of anxiety when presenting their speeches, and their speeches were independently judged to be superior in quality. Once more, this shows that when a single session of avoid-threat ABM achieves its intended objective of reducing attention to threat, it can attenuate the severity of previously existing anxiety dysfunction, subsequently assessed in the same laboratory session.

Single session ABM likewise can reduce anxiety reactivity to a laboratory stressor in people with elevated levels of general anxious disposition. Bernstein and Zvielli (2014) recruited participants who scored high in trait anxiety and displayed an attentional bias toward threat. These participants completed a single session of a probe task employing threatening and nonthreatening images. For half the participants this task was configured in the standard control condition, and for half it was configured in an intended avoid-threat ABM condition. This ABM condition was unusual, in that probes were shown with equal frequency in the locations of threat and nonthreat stimuli, and the researchers sought to influence attentional selectivity using principles akin to those that underpin biofeedback training. Specifically, a score was continuously computed, using probe latencies from a moving window of immediately prior trials, that indexed biased attentional selectivity, and was used to provide continuous real-time feedback on a display that showed a bar moving up and down a vertical scale. Participants were instructed to perform the probe task in a manner that kept the bar in the middle of this scale. This midpoint actually corresponded to unbiased attentional responding to threat. Bernstein and Zvielli were able to confirm that, relative to those who received the control condition, participants given this intended ABM training did indeed come to display a significant attenuation of their prior attentional bias to threat. This successful reduction of attentional bias was accompanied by the corresponding reduction of their heighted anxiety vulnerability, as revealed by briefer anxiety reactions to a subsequently presented anxiogenic video clip, for participants who received the avoid-threat ABM compared with those in the control condition.

Dandeneau and Baldwin (2009) investigated whether a single session of ABM could reduce the feelings of rejection elicited by a simulated social stressor in participants with low self-esteem. These researchers also employed a novel ABM approach. Rather than introducing a training contingency into the type of probe task that previously has been used to assess anxiety-linked attentional bias, Dandeneau and Baldwin (2004, 2009) instead introduced a training contingency into the type of visual search task previously used to assess such attentional bias. This intended ABM variant of the visual search task required participants to search through 112 grids of emotional expressive faces to find the single happy face among otherwise angry distracter faces in each grid. By requiring participants to repeatedly ignore angry faces and attend to happy faces, it was hoped that this task configuration would reduce attention to threat, and encourage attention to more positive information. A control version of the task, not intended to modify attentional responding to threat, instead required participants to search 112 grids for the single 5 petal flower among flowers that otherwise all had 7 petals. Using a conventional probe task to assess attentional selectivity, Dandeneau and Baldwin (2009) confirmed that, compared with the control condition, the intended ABM training reduced the degree to which participants with low self-esteem attended to rejection-related threat information. Furthermore, it influenced responses to a subsequent laboratory stressor, contrived to simulate a social rejection situation. Participants who had completed the search task in the avoid-threat ABM condition, compared with those who had completed it in the control condition, reported experiencing less intense feelings of rejection and fewer thoughts concerning rejection, in response to this social stressor.

Once again, spider fear has been the exception to the general finding that a single session of avoid-threat ABM training, when it successfully reduces attention to threat, serves to attenuate dysfunctional anxiety symptoms in the laboratory setting. Reese, McNally, Najmi, and Amir (2010) recruited participants who reported high levels of spider fear. All completed a single session of a probe task that employed pairs of images, and in each such pair one image was a spider (threat). Half the participants received this task in the avoid-threat ABM condition, whereas half received it in the standard control condition. Attentional assessment confirmed that this manipulation successfully induced the intended group difference in selective attentional responding to spider information. However, when subsequently required to walk toward a live spider, these two groups of participants did not differ either in the intensity of anxiety they experienced or in the degree to which they were able to approach the spider. Once more, this suggests that biased attention to spider-related information may not causally contribute to this particular manifestation of dysfunctional anxiety

Aside from the solitary exception of spider fear, the results of single session ABM studies offer strong support for the hypothesis that selective attentional responding to threat makes a causal contribution to anxiety variability and dysfunction, and lend encouragement to the exciting possibility that ABM approaches may be capable of delivering therapeutic benefits. However, demonstrating that a single session of ABM exerts an immediate impact on anxiety reactivity, to a contrived laboratory stressor, does not permit the inference that ABM will influence naturalistic emotional experience outside the laboratory. To address this issue, many researchers now have directly examined whether more extended exposure to intended ABM training, delivered across multiple sessions, can influence dispositional and dysfunction anxiety within real-world settings. We now turn to consider the findings that have been generated by these studies.

Impact of extended ABM on anxiety when it has successfully trained attentional change

See, MacLeod, and Bridle (2009) investigated whether repeated exposure to the probe ABM training procedure across multiple sessions could attenuate real-world anxiety responses to a stressful life event. They delivered 15 daily sessions of this task to Singaporean high school graduates immediately prior to their departure to commence tertiary studies in Australia. Half the participants received the task in the avoid-threat condition, whereas half received it in the standard control condition. Participants exposed to these alternative training conditions exhibited the intended differential change in attentional selectivity, whereas those in the avoid-threat condition exhibited a reduction in attention to threat information not shown by those in the control condition. This trained reduction in attention to threat was accompanied by an apparent increase in real-world emotional resilience. The degree to which participants exhibited elevated state and trait anxiety scores, on subsequent arrival in Australia, was significantly and substantially attenuated for those who had received the avoid-threat ABM training, compared with those in the control condition. These findings confirm that successful reduction of attention to threat information, through extended sessions of ABM training, can ameliorate naturalistic anxiety responses to stressful life events.

Other studies have examined whether multiple sessions of avoid-threat ABM training can alleviate problematic anxiety symptoms in participants who show an elevated dispositional tendency to experience such symptoms. Hazen, Vasey, and Schmidt (2009) delivered five sessions of this probe ABM procedure to students who tended to worry excessively. Half received the task in the avoid-threat condition, and half received the task in standard control condition. Significantly different attentional change was observed, with the former participants alone displaying reduced attention to threat across this period. The two conditions also differed in terms of their impact on emotion. Participants who received the ABM training that served to successfully reduce attention to threat also experienced a significant decline in emotional dysphoria, not experienced by those in the control condition.

Li, Tan, Qian, and Liu (2008) took a similar approach to investigate whether such extended ABM training would reduce social anxiety symptoms in students scoring high on a measure of social anxiety. They completed 7 consecutive daily sessions of a probe task that employed as stimuli pairs of faces, one displaying a threatening and the other a neutral expression. For half the participants, trials were presented in the avoid negative ABM condition, whereas the remaining participants received the task in standard control condition. Li et al. confirmed that these two conditions successfully induced significantly different change in attentional selectivity as intended, with participants in the avoid negative ABM training condition alone coming to display attentional avoidance of the threat images. This successful attentional training also affected the participants’ social anxiety symptoms, which significantly declined across these 7 days in participants who received the ABM training, while remaining unchanged in those who received the control condition.

In none of the earlier-described studies were participants selected on the basis of clinical dysfunction. Perhaps the most exciting and influential extended ABM training studies have been those designed to evaluate whether training that reduces selective attention to threat can therapeutically attenuate symptom severity in people with clinical anxiety disorders. When extended exposure to intended ABM successfully exerts this intended attentional change, then so too does it produce such a therapeutic benefit. Amir, Beard, Burns, and Bomyea (2009) carried out such a study on treatment seeking patients with GAD, who all completed 8 biweekly sessions of the attentional probe task employing emotional word pairs. Half were given the task in the avoid-threat ABM training condition, and half were given it in the standard control condition. Again, this training manipulation proved successful in inducing different attentional change, with the former participants alone demonstrating a significant reduction in attention to threat across these 4 weeks. The avoid-threat ABM training also exerted a dramatic impact on clinical symptoms, producing a significant reduction in both self-report and interview-based measures of anxiety, which was not observed in the control condition. At the conclusion of the study, 50% of the participants who had received avoid-threat ABM training no longer met diagnostic criteria for generalized anxiety disorder, compared with only 13% of participants in the control condition.

Similar clinical benefits of ABM training that have proven successful in reducing attention to threat, have also been observed for patients with generalized social phobia (GSP). Amir, Beard, Taylor, et al. (2009) conducted a study similar to that just described, but delivering the 8 biweekly sessions to individuals meeting diagnostic criteria for GSP. The probe ABM task used emotionally expressive faces as stimuli, rather than emotional words, to maximize social relevance. As before, half the participants received the task in the avoid-threat condition, where probes always appeared distally to the more threatening faces, whereas half received it in the standard control condition. Again, this manipulation achieved the intended attentional effect, with participants given the avoid-threat ABM training showing a significant reduction in attention to social threat that was not evident in those who received the control condition. Once more, this differential attentional change was accompanied by a differential change in anxiety. Specifically, participants who received avoid-threat ABM training showed significantly greater reductions than those in the control condition on all questionnaire measures of social anxiety. At the end of the 4-week study, 50% of participants given avoid-threat ABM training no longer met diagnostic criteria for GSP, which was the case for only 14% of those in the control condition. Symptom improvement was maintained at 4-month follow-up.

Several studies have shown that when ABM training successfully reduces attention to threat it also alleviates symptom severity in anxious children. Bar-Haim, Morag, and Glickman (2011) demonstrated this in 10-year-olds with dysfunctional anxiety. They completed two sessions of a probe task employing emotional faces, in a two week period, configured either in the avoid-threat ABM training condition or in the standard control condition. This manipulation was successful in inducing significantly different attentional change, with children who received avoid-threat ABM training subsequently displaying reduced attention to threat faces, compared with children in the control group. Several days later all children were exposed to a stressful puzzle task, and the two groups differed significantly in their emotional reaction. Although children in the control condition exhibited a pronounced anxiety elevation in response to this stressor, those who had received the ABM training shown to reduce attention to threat displayed no such elevation of anxiety.

Eldar et al. (2012) subsequently conducted a similar study with clinically anxious children aged 8 to 14 years. These children were given four weekly sessions of a probe task, again using emotional faces as stimuli, configured in either the avoid-threat ABM training condition or the standard control condition. This training was successful in altering attentional selectivity as intended. Children who received the avoid-threat ABM training showed a reduction in attention to threat faces that was not observed in children who received the control condition. Once again, having served to attenuate attention to threat, this ABM training also proved to attenuate clinical symptom severity. Anxiety symptoms were significantly reduced for children who received the avoid-threat ABM training, but did not change for children in the control condition. At the conclusion of the study, 33% of the former children no longer met diagnostic criteria for an anxiety disorder, compared with only 13% of the latter children.

Positive findings also have been obtained from studies that have used multiple sessions of the visual search ABM task variant to train reduced attention to threat in clinically anxious children. Waters, Pittaway, Mogg, Bradley, and Pine (2013) gave ten sessions of this task, across three weeks, to clinically anxious children aged 7 to 13 years. Half completed it in an avoid-threat ABM training condition that required them to search for a single happy face in a grid of angry faces. Half completed a control condition that required them to search for a single bird image in a grid of flower images. A face version of the probe assessment task confirmed that children who received the avoid-threat ABM training, but not those in the control condition, came to show a decrease in attention to threatening relative to positive faces. This ABM induced change in attentional selectivity was accompanied by a corresponding change in severity of anxiety symptoms, which decreased significantly over this period for the children given the avoid negative ABM training, but did not change for those in the control condition. Of the children who received avoid-negative ABM training, 33% no longer met diagnostic criteria for an anxiety disorder, at the conclusion of the study, compared with only 6% of those in the control condition. De Voogd, Wiers, Prins, and Salemink (2014) have replicated these positive findings in 13- to 16-year-olds reporting social phobia. They completed just two sessions of this visual search task in a 7-day period, with half receiving the avoid-threat ABM training condition and half the control condition. Attentional change differed significantly for these subgroups of children, with the former alone coming to display a more positive attentional bias across this week. This attentional change was accompanied by a significant reduction of social phobia symptoms, whereas no symptom change was apparent in children who received the control condition.

Across all these studies, in which extended delivery of multiple ABM training sessions has successfully attenuated attention to threat information, then so too has this attention change been accompanied by the attenuation of real-world anxiety. It has increased emotional resilience and reduced dysfunctional anxiety symptoms in nonclinical participants. It also has alleviated clinical symptom severity in people suffering from anxiety disorders. If these emotional benefits are truly caused by this training induced change in attentional selectivity, then they should not be observed when intended ABM procedures fail to elicit the target attention change. Before considering the emotional impact of intended ABM training procedures that have demonstrably failed to modify attentional bias, we first will briefly review studies where the attentional impact of intended ABM training procedures has been uncertain.

Impact of extended ABM on anxiety when its impact on attention has been uncertain

Some studies have evaluated whether multiple sessions of intended ABM training affects participants’ anxiety in real-world settings, without verifying that the procedures successfully modify these participants’ attention as required. Most such studies have found that the intended ABM training therapeutically alleviates anxiety, and the single exception to this can readily be explained by a possible failure to successfully train the requisite attentional change.

Having established that a single session of their visual search ABM task modifies attentional selectivity in a previous participant sample, Dandeneau, Baldwin, Baccus, Sakellaropoulo, and Pruessner (2007) went on to examine the emotional impact of delivering two sessions of this task to people confronting stressful situations, without measuring its impact on attention. Students approaching examinations and telemarketers in a stressful work environment were given the search task either in the avoid-threat ABM training condition or in the control condition. The former condition, compared with the latter, resulted in significantly increased self-esteem and reduced anxiety, as assessed by self-report, observer-report, and physiological measures.

A study by Schmidt, Richey, Buckner, and Timpano (2009) gave patients with social anxiety disorder (SAD) the same probe ABM procedure employed by Amir, Beard, Taylor, et al. (2009), assessing only its emotional impact without measuring attentional change. After 8 sessions, delivered across 4 weeks, patients who received the task in the avoid-threat condition displayed significant reductions in both self-report and clinician-rated measures of anxiety, not observed for patients in the control condition. At the end of this period, 72% of the former participants no longer met diagnostic criteria for SAD, compared with 13% of the latter. These impressive therapeutic gains were generally maintained at four month follow-up. In a more recent study, Riemann, Kuckertz, Rozenman, Weersing, and Amir (2013) delivered these same two conditions, again in eight sessions over 4 weeks, to clinically anxious 13- to 17-year-olds who also were receiving conventional CBT. Again, no measure of attentional change was taken, but those who received the avoid-threat ABM condition displayed significantly greater reduction of anxiety symptoms than did those in the control condition. Clinically significant improvement was achieved in 52.4% of the former participants, compared with only 4.8% of those in the control condition.

In contrast, when Bunnell, Beidel, and Mesa (2013) gave this same intended ABM training procedure and control condition to adults meeting diagnostic criteria for SAD, they observed no differential change in anxiety symptoms. Unfortunately, they too neglected to assess attentional selectivity. Hence it may be that a manipulation failure, reflecting unsuccessful training of attentional change in this particular study, may explain the absence of a change in anxiety symptoms. We advocate that future studies of this type should always include a measure of attentional change, as is necessary to verify whether an observed failure to induce change in anxiety using an intended ABM training procedure reflects the failure of this procedure to successfully modify attentional bias.

Another reason why the success of an intended training procedure in modifying attentional bias can remain indeterminate is inadequate statistical power. This limitation compromises confident conclusions concerning attentional change in two case study series. Bechor et al. (2014) carried out one of these using only 6 clinically anxious children, aged 10 to 13 years. Children were given the avoid-threat ABM procedure used by Amir, Beard, Taylor, et al. (2009), and across the 4-week period mean reduction on all the self-report measures of anxiety was significant. The nominal reduction in attention to threatening information did not achieve significance, but the study was inadequately powered to detect such an effect statistically. Rozenman, Weersing, and Amir (2011) report a case series conducted on 16 clinically anxious youth aged 10 to 17 years. Across 4 weeks all received 12 sessions of this same intended ABM procedure. There was a statistically significant reduction in self-rated, parent-rated, and clinician-rated anxiety symptoms, which was clinically significant for 10 of the 16 participants. Again, there was a nominal reduction in attention to threat information, but this fell outside statistical significance, which the investigators attribute to small sample size. Nevertheless, across participants the magnitude of reduction in attention to threat significantly predicted reduction in clinician-rated anxiety.

Thus, when the attentional impact of extended ABM training intended to reduce attention to threat has remained uncertain, it has usually, but not universally, reduced real-world anxiety symptoms. Our present suggestion that the occasional absence of such emotional benefits, in these studies, may reflect their occasional failure to successfully manipulate attention is lent further weight by the emotional outcomes of studies in which intended ABM training has clearly failed to modify attentional selectivity.

Impact of extended ABM on anxiety when it has failed to train attentional change

When attentional selectivity has been appropriately assessed, and an intended ABM training procedure has demonstrably failed to successfully modify attention to threat, then the procedure has exerted no impact on anxiety.

For example, Maoz, Abend, Fox, Pine, and Bar-Haim (2013) gave socially anxious students four sessions of a probe task training procedure, using emotional face pairs, configured either in the avoid-threat or standard control condition. But Maoz et al. presented the stimuli subliminally, and this novel task variant did not serve as an ABM procedure, as there was no tendency for the two conditions to differentially influence attention to threat. Corresponding, there was no tendency for the two conditions to differentially influence self-reported anxiety, or anxiety reactivity to a final stressor. These findings are consistent with our contention that intended ABM training procedures only attenuate anxiety if they successfully attenuate attention to threat.

Enock, Hoffman, and McNally (2014) took the innovative step of delivering an intended ABM procedure, similar to the probe approach employed by Amir, Beard, Taylor, et al. (2009), on a smartphone platform. Participants completed this task on their own mobile phones 3 times each day across 4 weeks. For one group of participants the task was configured in the avoid-threat condition, whereas for the other participants it was configured in the standard control condition. An attentional assessment task revealed no differential change in selective attentional responding to threat information between these two groups of participants, indicating that the intended ABM manipulation was unsuccessful in reducing attention to threat. Likewise, there was no differential change in anxiety symptoms, supporting our premise that successful ABM is necessary to influence anxiety symptoms. Possibly the small and highly variable screen size may have compromised the intended attentional training manipulation, or perhaps this manipulation failure was due to participants commonly completing the intended ABM training task on their phones under more distracting and disruptive circumstances than is the case when ABM training is delivered in the controlled laboratory setting.

Consistent with this latter possibility, when researchers have remotely delivered intended ABM procedures, previously shown to effectively train attentional change within the laboratory, to participants in their home environments, these procedures have commonly failed to modify attentional bias (Boettcher, Berger, & Renneberg, 2012; Boettcher, Hasselrot, Sund, Andersson, & Carlbring, 2014; Boettcher et al., 2013; Carlbring et al., 2012; Rapee et al., 2013). In each of these studies, participants have been people with SAD. The home-delivered procedures have been probe tasks using either expressive face pairs or emotional word pairs as stimuli, and the studies have contrasted an intended avoid-threat ABM condition against the standard control condition. Participants have completed these tasks at home using their own computers, across periods ranging from 2 to 12 weeks, typically without any constraint on the nature of the setting within which task completion takes place. In each study, pre and post measures of attentional selectivity have consistently revealed that the intended ABM training condition did not reduce attention to threat. And, without exception, this absence of differential change in attentional responding to threat, has been accompanied by the absence of differential change in anxiety symptoms. These home-delivered protocols might fail to operate effectively as ABM procedures because participants may carry out the home based sessions in the presence of multiple distractions, from television, music, family conversation, and so forth. As we will later discuss, it will be important for future researchers to address such possibilities, given the potential benefits of being able to successfully train ABM in the home setting. For the moment, these studies confirm that, unless an intended ABM training procedure designed with the aim of attenuating attention to threat actually produces such ABM, it does not serve to attenuate anxiety.

Occasionally, researchers have failed to achieve intended ABM within the laboratory setting, using procedures that have previously proven successful in producing such attentional change. When this has been the case, then these procedures also have produced no change in anxiety either. Across 4 weeks, McNally, Enock, Tsai, and Tousian (2013) gave speech anxious participants four sessions of a probe task employing emotional face pairs, configured either in the avoid-threat or standard control condition. Measures of attentional selectivity revealed that participants completing the tasks in the alternative conditions showed no differential change in attention to threat. Likewise, they showed no differential change in self-report, behavioral, or physiological measures of anxiety. Britton et al. (2013) and Shechner et al. (2014) gave similar probe tasks, again using facial stimuli, to clinically anxious patients, across 8 and 16 sessions, respectively. Measures of attentional selectivity revealed no differential change in attention to threat between participants who received the tasks in the intended avoid-threat condition, and those who received it in the standard control condition. Correspondingly, neither study revealed evidence of differing change in anxiety symptoms between the participants exposed to these alternative conditions. Finally, Schoorl and colleagues (Schoorl, Putman, & Van der Does, 2013; Schoorl, Putman, Mooren, Van der Werff, & Van der Does, 2014) have delivered probe tasks of this nature, across multiple sessions, to patients with posttraumatic stress disorder (PTSD). The task variants configured in the intended avoid-threat condition, and in the standard control condition did not differentially modify attentional responding to threat. Again, in the absence of the intended modification of attention to threat, there was no attenuation of PTSD symptom severity.

Illuminating the basis of individual differences in attentional bias malleability

Meta-analyses that have examined the association between CBM in general (rather than ABM specifically) and composite measures of symptom change (rather than anxiety change specifically) sometimes have observed no correlation between cross-study variability in magnitude of CBM-induced cognitive change and cross-study variability in symptom improvement (Hallion & Ruscio, 2011; Mogoase et al., 2014). However, in Hakamata et al.’s (2010) meta-analysis, that specifically examined the impact of ABM on anxiety, induced change in attentional bias was found to correlate .75 with observed change in anxiety across studies. Moreover, many (though not all) ABM studies have found cross-participant variability in the magnitude of ABM-induced attentional change to be significantly related to cross-participant variability in the resulting change in anxiety (e.g., Amir, Beard, Burns, et al., 2009; Amir, Beard, Taylor, et al., 2009; Bar-Haim et al., 2011; MacLeod et al., 2002; Najmi & Amir, 2010; Rozenman et al., 2011; See et al., 2009). This may reflect a continuous association between degree of induced attention change and degree of resulting anxiety change, or it could result instead from the fact that in the absence of attentional change no such anxiety change occurs. If the latter is the case, then the strength of this observed association will differ across studies, being most evident when there is high variability in terms of whether or not significant attentional change is achieved across participants, and least evident when ABM is successful in inducing significant attentional change across all participants. Whatever its basis, it is clearly important to understand what gives rise to variability in the attentional impact of intended ABM manipulations.

Some participants are more likely than others to display substantial attentional change in response to intended ABM training procedures, and there is now compelling evidence that this variability partly reflects a systematic individual difference in attentional bias malleability. Clarke, MacLeod, and Shirazee (2008) exposed newly commencing university students to a brief probe ABM task in the attend-threat condition, and measured the degree to which each participant evidenced the intended transient change in attentional bias, to yield an index of attentional bias malleability. This early measure of attentional bias malleability predicted the degree to which these students subsequently responded to the extended mild stress of their first university semester by naturalistically developing increased attention to threat, as assessed on the final week of semester. Individuals who initially demonstrated greater attentional bias malleability not only came to develop heighted attention to threat across the subsequent semester, but also displayed a greater increase in the frequency of anxiety symptoms across the course of this semester, an emotional change that was mediated by the concurrent change in attentional bias. More recently, Clarke, Chen, and Guastella (2012) found that a similar initial measure of attentional bias malleability, taken in treatment-seeking patients with SAD, also significantly predicted the magnitude of their subsequent therapeutic response to conventional cognitive behavior therapy. Patients with higher levels of attentional bias malleability displayed a greater therapeutic response to this intervention in terms of anxiety reduction. Thus it seems plausible that individual difference in the malleability of attentional bias may contribute to variation in the malleability of anxious symptomatology.

We believe future researchers should now seek to identify the cognitive mechanisms that underpin this individual difference in the degree to which intended ABM training induces the target attentional change. Minimally, this may permit screening to identify candidate recipients most likely to therapeutically benefit from ABM training, by acquiring the change in attentional bias necessary to alleviate dysfunctional anxiety. It also is likely to guide development of future ABM training variants that more consistently induce, across all recipients, this required attentional change. Fox, Zougkou, Ridgewell, and Garner (2011) have shown that individual differences in attentional bias malleability partly reflects genetic factors. These researchers found that attentional bias malleability, indexed by the magnitude of change in attentional bias induced by a single session of probe ABM training, was significantly greater in individuals carrying the short variant of 5-HTTLPR, a polymorphism of the serotonin transporter gene associated with the amplified emotional impact of both negative and positive environmental experiences (cf. Belsky & Hartman, 2014). Although these findings confirm that systematic individual differences moderate the attentional impact of intended ABM training procedures, they do not reveal the nature of cognitive phenotype that gives rise to heightened attentional bias malleability.

One possibility is that heightened attentional bias malleability may reflect an elevated tendency to register the emotional valence of stimulus information. Unless participants are sensitive to the differing emotional valence of stimulus information, then intended ABM training contingencies cannot exert an influence on attentional selectivity. If this valence sensitivity hypothesis is valid, then ABM training variants designed to ensure effective registration of stimulus valence might yield more robust and consistent attentional change. Accordingly, some researchers have taken steps to encourage valence sensitivity in their ABM protocols. In a recent probe ABM training study, Grafton, Ang, and MacLeod (2012) interspersed training trials with valence judgment trials. On these latter trials no probe appeared, but instead participants were required to classify the valence of one of the preceding words, to encourage them to encode stimulus emotionality. The intended pattern of attentional selectivity was successfully induced by this amended training procedure, which also significantly influenced emotional reactivity to subsequent anagram task. However it remains unknown whether the ABM training was significantly enhanced by the presence of these valence judgment trials. Future research should investigate whether individual differences in valence sensitivity predict attentional bias malleability, and whether the impact of intended ABM training on attentional selectivity can be significantly increased by procedures designed to increase cognitive registration of stimulus valence.

Another plausible hypothesis is that individual differences in executive functioning may contribute to variation in attentional bias malleability. ABM training procedures are essentially learning tasks, and individual differences in executive functioning can moderate efficacy of learning (cf. Titz & Karbach, 2014). Browning, Holmes, Murphy, Goodwin, and Harmer (2010) have reported neurocognitive evidence that executive functioning is involved in successful ABM training. Specifically, these investigators demonstrated that participants given probe ABM training that successfully modified attentional responding to negative information displayed increased lateral prefrontal cortex (lPFC) activation when processing information of the valence they were trained to avoid. Given that lPFC activation is associated with executive functioning (cf. Stuss & Knight, 2013), this suggests that higher levels of executive functioning may enhance attentional bias malleability. In a direct test of this hypothesis, Clarke, Browning, Hammond, Notebaert, and MacLeod (2014) delivered transcranial direct current stimulation (tDCS) to increase lPFC activation in a sample of participants completing a probe ABM training task. These participants not only displayed the intended change in attentional selectivity, but did so to a significantly greater degree than did participants a sham tDCS condition that delivered no current. The enhancement of attentional responsiveness to ABM training through lPFC stimulation supports the idea that executive functioning contributes to attentional bias malleability, while also indicating that the judicious use of frontal tDCS potentially may further improve ABM outcomes.

Many other factors, such as individual differences in the registration of basic frequency information, or in the detection of covariation, also may contribute to individual differences in attentional bias malleability. Future research that illuminates the nature of these cognitive variables, and exploits the resulting knowledge to improve prediction concerning who will benefit from ABM training, and to increase its capacity to modify attentional selectivity, is likely to pay valuable dividends.

Enhancing the delivery of existing ABM training tasks

It appears that some methods of delivering existing ABM training procedures may be more effective than others. Hence, future research informing how to most effectively deploy these training procedures will clearly be important. On the basis of meta-analysis, Hakamata et al. (2010) showed that ABM studies using word stimuli have trained greater change in attentional selectivity than those using pictorial stimuli, and those presenting stimulus pair members aligned vertically have trained greater attentional change than those in which they have been aligned horizontally. However, reliance on meta-analysis alone to determine the most effective ABM delivery format is limited by two factors. First, across existing studies, task parameters often have been confounded. For example, ABM studies using word pairs usually have aligned stimuli vertically, with horizontal alignment being more common when stimuli have been images. Also ABM training using image stimuli has been more common with certain participant populations than with others, meaning that stimulus variation has been confounded with participant variation. A second limitation of reliance on meta-analysis is that this approach can only consider the influence of parameters that already have commonly varied across prior ABM studies. It may be that ABM training procedures can be enhanced by delivering them in ways that differ from most previous approaches.

For example, it is possible that altering participant instructions might further increase ABM efficacy. In virtually all ABM training studies carried out to date using the probe approach, participants have not been told that probe position is predicted by threat position, perhaps reflecting the untested assumption that ABM will be most effective when implicit learning drives attentional change. The results of two studies designed to examine the veracity of this assumption have been somewhat inconsistent, but both suggest that the instructions given to participants can influence the impact of ABM training. Grafton, Mackintosh, Vujic, and MacLeod (2014) gave student participants a single session of a probe ABM procedure, configured in either the avoid-threat or attend-threat training condition, and examined its impact on both attentional selectivity and anxiety reactivity to a stressor. For some participants instructions made no mention of the training contingency, in keeping with most previous ABM studies. For these participants, the ABM manipulation induced the intended differential pattern of attentional selectivity, and also influenced anxiety reactivity to the stressor, thereby replicating previous findings. For another sample, the instructions explicitly informed participants where probes would appear relative to threat and directly instructed them to actively practice adopting the target pattern of attentional selectivity by moving attention to the anticipated locus of the probe. For these participants, the impact of ABM condition on attentional selectivity was significant, and especially pronounced, with an effect size twice the size of that observed in the absence of such explicit instruction. However, there was now no impact of this attentional manipulation on anxiety reactivity to the stressor. Grafton et al. suggest that instructing adoption of the target attentional selectivity during probe task performance may have produced an attentional change that did not generalize and endure beyond such task performance, as would be necessary to influence anxiety responses to subsequent situations.

A study by Krebs et al. (2010) investigated the impact of a slightly different manipulation on ABM efficacy. Krebs et al. gave a single session of the probe ABM task, in the same two training conditions used by Grafton et al. Some participants were given conventional instructions. Others were told where the probes would appear relative to threat, though Krebs et al. did not explicitly direct participants to actively adopt and practice the target pattern of attentional selectivity. The ABM training manipulation was successful in eliciting the intended difference in attentional responding to threat, and this effect was significantly greater for those participants who had been explicitly informed about the training contingency. Krebs et al. then told participants to engage in intentional worry, and examined the degree to which this elicited subsequent negative thought intrusions. Participants given the avoid-threat ABM condition reported less negative thought intrusions following this directed worry than did those given the attend-threat ABM condition, and this effect was significantly more pronounced when explicit information about the ABM training contingency had been provided in the instructions. Whether the discrepant emotional consequences of the explicit instruction condition in these two studies reflects the differing nature of their specific instructional manipulation, or the differing anxiety-related measure employed, remains uncertain for the present. What seems clear, however, is that the precise wording of instructions has the capacity to affect the outcomes of intended ABM training procedures. Hence, the beneficial outcomes of ABM could likely be increased by future research that identifies the optimal way of framing instructions.

Even when ABM tasks have employed equivalent instructions, it has commonly been found that multiple sessions delivered remotely to the home setting often fail to produce the intended attentional change, that more reliably results from their delivery within the controlled laboratory setting. It is important for future research to confront this discrepancy. A conservative approach may advocate in favor of restricting delivery of ABM training procedures to the laboratory or clinic, given that this is the context within which it presently produces the most reliable attentional change, and so yields the best therapeutic outcomes. A more ambitious approach will be to identify the reasons why attentional change has proven difficult to reliably induce in the home setting and will aim to resolve these problems. As we have pointed out, one possibility is that home-based sessions are more likely to be completed under distracting conditions, with interference from competing activities reducing their efficacy in modifying attention. This could be tested by comparing existing home-delivery procedures against variants designed to more closely emulate laboratory conditions. These may require participants to complete the home-based training in a quiet setting, alone and without the prospect of interruption. If such straightforward refinements enable home-delivered ABM to reliably train the attentional change it commonly produces within the laboratory, with the emotional benefits that typically accompany such attentional change, this could greatly increase its accessibility, and hence its therapeutic value.

Kuckertz et al. (2014) recently proposed an alternative candidate explanation for the common failure of home-delivered ABM procedures to produce intended attentional change. They note that participants may commonly feel more anxious in the laboratory than they do at home, and so suggest that intended ABM training procedures may most effectively modify attentional selectivity when completed while participants are experiencing elevated anxiety. These investigators replicated a previous study carried out by Carlbring et al. (2012), in which a home-delivered intended ABM procedure failed to reduce attention to threat in participants with SAD. However, Kuckertz et al. now told participants that immediately before completing each home session they should “challenge themselves with a social anxiety oriented task” such as placing a difficult phone call, to elevate their anxiety. This amended home-based ABM training now significantly reduced attention to threat as intended, and this attentional change was accompanied by a significant reduction in social anxiety symptoms, equivalent in magnitude to the therapeutic improvement resulting from a cognitive behavior therapy comparison condition. Although Kuckertz et al.’s findings lend weight to their hypothesis that intended ABM procedures may more effectively modify attentional selectivity when participants are experiencing elevated anxiety, alternative explanations cannot yet be discounted. It is possible that fulfilling the instruction to perform an anxiety-inducing act, such as placing a difficult phone call, immediately before each home-delivered ABM session, may have led participants to remove themselves from other distractions, that otherwise would have compromised ABM. Future research should seek to identify the specific mechanism through which Kuckertz et al.’s amended procedure enhanced the efficacy of home-delivered ABM.

Many other aspects of delivery potentially could affect the capacity of presently used ABM tasks to successfully induce and sustain the intended change in attentional selectivity. These include individual session length, the number of such sessions, and their precise scheduling. For example, on the basis of learning literature Hertel and Mathews (2011) have suggested that the magnitude of induced attentional change may be maximized by delivering multiple training sessions in close succession, whereas the retention of induced attention change may maximized by increasing the temporal separation of training sessions. Research that systematically delineates how differences in their delivery influence the degree to which established ABM procedures achieve intended attentional change will likely be of great value in guiding the development of delivery protocols that optimize their benefits.

Development of new and more effective ABM training tasks

Although enhancing the delivery of existing ABM tasks is important, future research should not be blinkered to the possibility that entirely new types of attentional training procedures may prove superior to those in current use. The great majority of ABM studies to date have employed probe task variants, configured with the intention of inducing a target pattern of ABM. Indeed, at times researchers have appeared to confuse this specific task with the concept of ABM itself, claiming to have carried out ABM when they have delivered a probe task of this nature, regardless of whether or not it has served to modify attentional bias. This intended training task cannot be considered an ABM procedure when it fails to modify attentional bias. Conversely, any quite different cognitive training task that does successfully modify attentional bias should be considered an ABM procedure.

We consider it improbable that the probe approach represents the best possible method of modifying attentional bias, or of assessing the attentional change that results from intended ABM procedures. The dot probe task is known to have low internal reliability (Schmukle, 2005). This has not prevented the task from differentiating the patterns of attentional selectively demonstrated by cohorts of high and low anxious individuals (cf. Bar-Haim et al., 2007), or by participant cohorts exposed to alternative ABM training conditions (cf. Hakamata et al., 2010), because such group comparisons aggregate assessment across all individuals in these comparison cohorts. However, low reliability will likely compromise its capacity to sensitively assess attentional bias at the individual participant level. Future researchers must look beyond this task, to develop new methods of ABM training that overcome its limitations, and more powerfully modify attentional selectivity.

Aside from the low internal reliability of the probe task, the monotonous nature of the probe-based ABM approach can lead to boredom, fatigue and loss of concentration, especially across lengthy periods of training. In many areas of cognitive training, concerns of this type have motivated the “gamification” of training procedures, to increase participant engagement in ways that may improve efficacy (cf. Kapp, 2012). Developing and evaluating new, more game-like approaches to ABM training also is likely to prove profitable. Some changes of this nature may not require major alteration to the stimulus displays used in previous probe tasks. For example, Notebaert, Clarke, Grafton, and MacLeod (under review) have reported a gamified ABM task variant, based on the children’s card game snap, which presents successive “cards” that each display pairs of faces, one showing a negative and one a positive expression, as in many probe ABM tasks. Participants play the game by quickly deciding whether each successive card shows the same image as the card before, as in the conventional snap game. However, half the participants were required to indicate whether the people expressing positive emotion on successive cards matched in identity, whereas half instead were required to indicate whether the people expressing negative emotion on successive cards matched in identity. Using a conventional probe assessment task, Notebaert et al. found that after completing 16 games, each presenting 24 cards, which took around 30 min, participants who had played it in the former condition showed greater attentional avoidance threat than did those who had played it in the latter condition. The former group of participants also displayed reduced anxiety reactivity to a final stressor, relative to the latter. Moreover, a single session playing this emotional snap game exerted significantly greater attentional and emotional impact than did a length matched session of the widely used probe ABM training procedure. It remains to be determined, by future research, whether individuals with SAD who play this relatively enjoyable game variant of avoid-threat ABM training across a more extended period, perhaps interactively in a social setting, will show greater therapeutic benefits than those who instead are given avoid-threat training using the probe ABM task.

Although gamified versions of ABM tasks can be simple, they also could potentially be more complex than existing ABM procedures, and complexity may bring additional benefits. The commonly used probe ABM task presents two static stimuli and requires participants to perform a very easy probe discrimination. This seems unlikely to challenge the attentional system to a degree that strongly encourages adoption of the intended attentional selectivity to satisfy the performance demands of the task. The visual search variant of ABM, which requires participants to search for a target emotional face in a small grid of static unchanging faces, represents only a small improvement in this regard. Real-world attentional selectivity must operate within informational environments that are complex and dynamic. Grafton, Notebaert, and MacLeod (2013) have reported a gamified version of ABM training that presents emotional information in a more complex and dynamic manner. In this game, multiple faces all swiftly and continuously moved around a computer screen, and participants tracked the single face displaying the target emotion, using the mouse cursor. When this face ceased to express this target emotion, then participants had to quickly switch to tracking the different face that now displayed this emotion. Game points accrued from accurate tracking and, across games, the participant’s “high score” indicated the performance they should strive to beat. Participants completed six games across 40 min, either in an avoid-threat ABM condition which required tracking the single happy face in a background of many angry faces, or in an attend-threat ABM condition, which required tracking the sole angry face among many happy faces. Participants who completed the games in these two conditions came to display the intended differential patterns of selective attentional responding to threat. They also differed significantly in their anxiety reactivity to a final stressor, which were substantially attenuated for participants who had received the avoid-threat condition.

In another recent gamified ABM variant, delivered on a smartphone platform to students with high trait anxiety, Dennis and O’Toole (2014) presented two animated cartoon characters, moving through grass, one showing an angry and the other a happy expression. When both burrowed into the grass, a “trail” remained visible in one of these locations, which participants had to quickly touch. Points accrued for quick and accurate responding, with “jewels” being awarded for good performance. Half the participants completed the task in an avoid-threat ABM condition, intended to reduce attention to threat by consistently presenting trails in the opposite location from the character with the angry expression, and half received the task in a control condition, where trails appeared equally often in the two locations. Attentional selectivity was assessed using a conventional computer-delivered probe task that employed photographs of real faces. This confirmed that, following a single 45 min session playing this game, participants who completed it in the avoid-threat ABM condition demonstrated reduced attentional bias to threat, which was not the case for those who completed it in the control condition. Furthermore, in contrast to control participants, those who played the game in the avoid-threat ABM condition also evidenced significant reductions in both self-report and behavioral measures of anxiety reactivity to a final stressor.

We are optimistic that the near future will see the arrival of many such gamified variants of ABM tasks, enriching our available arsenal of candidate training techniques. It will be important not only to verify that these new tasks do exert the intended influence on attentional selectivity and anxiety, but also to directly compare their efficacy against that of existing ABM tasks. Such comparisons should examine the magnitude of induced change in attention and anxiety, the maintenance of this change across time, task completion rates, participant drop out, and user satisfaction. With professional programming input and ever improving technology there is no limit to the breadth and scope of potentially engaging and effective ABM procedures that can be developed, and we encourage future researchers to be fully open to these opportunities.