Journal of Affective Disorders 223 (2017) 175–183

Contents lists available at ScienceDirect

Journal of Affective Disorders

journal homepage: www.elsevier.com/locate/jad

Review article

Cognitive bias modification: A review of meta-analyses MARK

⁎
Emma B. Jones, Louise Sharpe
School of Psychology A18, The University of Sydney, 2006 NSW, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: Background: Cognitive bias modification (CBM) is a novel, but controversial intervention with considerable
Cognitive Bias Modification (CBM) divergence amongst conclusions in individual studies and reviews. This systematic review synthesizes meta-
Attentional Bias Modification (ABM) analyses of CBM to determine whether CBM is effective, and what parameters most reliably evoke the process of
Cognitive Bias Modification for Interpretation CBM.
(CBM-I)
Methods: A systematic literature search resulted in twelve meta-analyses in total, from which the published
Systematic review
effect sizes were extracted.
Results: Attention bias modification (ABM) shifted targeted biases in adults (ES = 0.24–1.16), was effective as a
buffer to stressor vulnerability (ES = 0.33–0.77) and in symptom control (ES = 0.16–0.41). Cognitive bias
modification for interpretation (CBM-I) modified targeted biases (ES = 0.52–0.81) but did not reliably reduce
stressor vulnerability (ES = 0.01–0.24, p > .05). CBM consistently reduced anxiety symptoms, but effects on
depressive symptomatology were less compelling. The long-term efficacy of CBM was only supported in ad-
diction studies.
Limitations: The review included a single CBM-I only meta-analysis, and two meta-analyses with pooled re-
porting on ABM and CBM-I outcomes.
Conclusions: Overall, this synthesis shows CBM is effective in the short-term for anxiety in adults, and highlights
some conditions under which CBM is most efficacious. Rather than debating the efficacy of CBM, future research
should focus on developing procedures that more reliably induce bias modification and determining the most
efficacious clinical applications.

1. Introduction replaces non-threatening stimuli during training. However, other

Cognitive bias modification (CBM1) refers to procedures that aim to
directly change automatic cognitive processes, such as attention and
interpretation that are hypothesized to contribute to the development
and maintenance of psychopathology. CBM is therefore an implicit
process, and is produced by directly modifying a targeted cognitive bias
in a particular direction (MacLeod and Mathews, 2012). Research has
focused primarily on two types of CBM: attention bias modification
(ABM) and cognitive bias modification for interpretation (CBM-I). In
ABM, participants are trained to attend to neutral or positive stimuli,
and avoid negative, threatening stimuli. Most commonly, the dot-probe
paradigm is used to modify biases by changing the contingencies be-
tween the target and the probe, such that the probe more frequently
paradigms have also been used for ABM including spatial cueing task
(e.g. Fox et al., 2001) and visual search (e.g. Dandeneau et al., 2007).
CBM-I, in contrast, requires participants to solve a task that dis-
ambiguates a sentence, paragraph, or picture to be either positively or
negatively valenced. Doing so leads participants to interpret new am-
biguous stimuli in the same manner that matches their training (posi-
tive or negative). The most widely used CBM-I training method is the
Ambiguous Scenarios (AS) paradigm (e.g. Blackwell and Holmes,
2010), while other training methods include the homograph paradigm
(e.g. Grey and Mathews, 2009) and the Word Sentence Association
Paradigm (WSAP) (e.g. Amir and Taylor, 2012).
Since the first ABM study published in 2002 (MacLeod et al.), the
number of studies of CBM has grown exponentially. The success of early

⁎
Corresponding author.
E-mail address: louise.sharpe@sydney.edu.au (L. Sharpe).
1
Approach Avoidance Task (AAT).
Attention Bias Modification (ABM).
Attentional Bias (AB).
Cognitive Bias Modification (CBM).
Cognitive Bias Modification for Interpretation (CBM-I).
Word Sentence Association Paradigm (WASP).

http://dx.doi.org/10.1016/j.jad.2017.07.034
Received 2 June 2017; Received in revised form 13 July 2017; Accepted 17 July 2017
Available online 18 July 2017
0165-0327/ Crown Copyright © 2017 Published by Elsevier B.V. All rights reserved.
E.B. Jones, L. Sharpe
experimental paradigms in inducing changes in emotional vulnerability
led researchers to see the potential therapeutic benefit of CBM, and as
such, recent studies are now targeting anxiety and depressive symptoms
in clinical settings.
As the literature on CBM has proliferated, so too has the number of
reviews and meta-analyses relating to CBM. The earliest meta-analysis
(Hakamata et al., 2010) concluded that ABM was a promising new
treatment with moderate to large effect sizes. However, some more
recent meta-analyses have suggested CBM is ineffective, or that its ef-
fects are so small that they are unlikely to be clinically relevant (Cristea
et al., 2016, 2015a, 2015b). It is possible that the varying conclusions of
the meta-analyses reflect a novel treatment that showed promise in
early, poorly controlled studies, but has not reached its early promise.
However, a cursory reading of the available meta-analyses reveals nu-
merous methodological differences between them that could, at least in
part, account for these discrepancies. It is therefore timely to provide a
narrative synthesis of available meta-analyses to determine whether the
source of differences can be determined.
The aim of this study is to review existing meta-analyses to identify
the variations in scope and methodology that give rise to the incon-
sistencies reported. Further, we aim to determine the conditions under
which CBM works most effectively. Hence, the research questions are:
(a) Does CBM work? And (b) If so, under what conditions is it most
effective?
2. Method
2.1. Literature search
All published meta-analyses on CBM were identified. Meta-analyses
had to include more than 4 CBM studies, and needed to meet PRISMA
guidelines. A systematic search of the literature was conducted in
PsychInfo, Medline, EMBASE and the Cochrane Library databases
through January 2017. The following key words were used from the
recent meta-analysis by Cristea et al. (2015a): "cognitive bias mod-
ification" or "attention* bias modification" or "attention bias training" or
"bias training" or "interpret* bias modification", which we then com-
bined with “meta-analysis”.
2.2. Study selection
The search process (see Fig. 1) resulted in 12 meta-analyses. All
articles were reviewed by the two authors, with perfect agreement
(Kappa = 1).
2.3. Summary of analyses
For a brief summary of the included analyses, see Table 1. The
current review focused exclusively on ABM and CBM-I studies. The
effects of CBM were examined directly using the following outcomes:
change in relevant bias, vulnerability to a stressor, and symptom
Fig. 1. PRISMA diagram.
176
Journal of Affective Disorders 223 (2017) 175–183
reduction. Post treatment outcomes that were reported in only one
meta-analysis were not included (n = 2). We also extracted moderating
variables examined in the meta-analyses. Reported effect size values are
those after outliers were removed. In this way, we have taken a con-
servative approach to the analysis. While we appreciate that the meta-
analyses differ in scope, since scope is not necessarily an indicator of
quality, per se, we will report the results and mention the scope of the
meta-analysis, where relevant, to the results.
3. Results
3.1. Do Cognitive bias modification protocols reliably change cognitive
biases?
3.1.1. Attention bias modification (ABM)
Nine out of the eleven meta-analyses that examined ABM measured
a change in attentional bias (AB). Neither Cristea et al. (2015a) nor
Kampmann et al. (2016) assessed change in biases. Of those nine meta-
analyses, eight found a significant effect of ABM on AB change (ES =
0.24–1.16). The remaining study used a child only population (Cristea
et al., 2015b). Hence, except for a meta-analysis exclusively with
children, all meta-analyses found that ABM protocols successfully
modified attention bias.
3.1.2. Cognitive bias modification for interpretation (CBM-I)
Only five of the twelve meta-analyses included CBM-I, and only
three examined change in interpretation bias. Nonetheless, in all three
meta-analyses, a significant effect of CBM-I (ES = 0.52–0.81) on in-
terpretation bias change was found overall. However, in comparing pre-
post positive bias change, Menne-Lothmann et al. (2014) found that the
effect of benign training only varied reliably from negative training (ES
= 0.65), not neutral or no training.
3.1.3. Follow-up outcomes
Mogoaşe et al. (2014) was the only meta-analysis to investigate the
degree to which bias change was enduring at follow-up. Changes in AB
were not maintained when measured between 2 weeks and 4 months
after the final ABM session.
3.2. Under what conditions do CBM protocols most reliably change biases?
3.2.1. Type of CBM intervention
Two analyses compared the relative success of ABM and CBM-I in
changing biases, with one study (Hallion and Ruscio, 2011) finding
CBM-I to be more successful, and the other (Cristea et al., 2015b)
finding no significant difference.
3.2.2. Age and sex
Of the five studies that assessed age as a moderator of CBM on bias
change, only one (Mogoaşe et al., 2014) revealed a significant effect.
Mogoaşe et al. (2014) found that younger participants benefited more
from ABM on bias change scores, but only in their anxiety subsample.
Similarly, mixed results were found in the three meta-analyses that
investigated the moderating role of gender. Menne-Lothmann et al.
(2014) revealed benign CBM-I to be more effective for females com-
pared with males, whereas Heeren et al. (2015) and Hakamata et al.
(2010) did not find significant effects.
3.2.3. Sample and psychopathology types
In the three studies that investigated type of pathology on AB
change, there was little evidence to support a relationship. Beard et al.
(2012) did find type of pathology to moderate AB change, but only for
training to neutral versus control; and this result was no longer sig-
nificant when two studies of smoking with small effects were removed.
Sample type as a moderator was examined in six studies, with 2/6
having significant findings. Both Beard et al. (2012) and Menne-
E.B. Jones, L. Sharpe Journal of Affective Disorders 223 (2017) 175–183

Table 1
Summary of included meta-analyses.

Study CBM type Age Sample type Focus

Beard et al. (2012) ABM Mostly adults Mix Clinically relevant outcomes
Cristea et al. (2016) ABM, AAT & Inhibition Adults Mix Substance addictions
Cristea et al. (2015a) ABM & CBM-I Adults Mix Clinically relevant outcomes
Cristea et al. (2015b) ABM & CBM-I Children Mix Clinically relevant outcomes
Hakamata et al. (2010) ABM Adults Mix Anxiety outcomes
Hallion and Ruscio (2011) ABM & CBM-I Adults Mix Clinically relevant outcomes
Heeren et al. (2015) ABM Adults Mix Social anxiety outcomes
Kampmann et al. (2016) ABM, AAT, AIM & CBM-I Adults SAD Dx Social anxiety outcomes
Linetzky et al. (2015) ABM Mostly adults Anxiety Dx Anxiety outcomes
Menne-Lothmann et al. (2014) CBM-I Adults Mix Negative mood/emotional reactivity
Mogoase et al. (2014) ABM Mostly adults Mix Clinically relevant outcomes
Price et al. (2016) ABM Mostly adults High anxiety Anxiety outcomes

ABM: attentional bias modification; AAT: approach-avoidance task; AIM: attention interpretation modification; CBM-I: cognitive bias modification for interpretation; Dx: diagnosis; Mix:
samples of diagnosed, analogue and healthy individuals; SAD: social anxiety disorder

Lothmann et al. (2014) found high symptomatology samples produced
a larger effect size than healthy samples, although only for positive vs
control in Beard et al.’s (2012) meta-analysis.
3.2.4. Protocol parameters
Two meta-analyses compared the effect of CBM on biases when
using pictures versus words. Both found stimulus modality to be a
significant moderator of AB change, but in Beard et al. the pictures
produced more shift, whereas in Hakamata et al. words produced
greater change. Menne-Lothmann et al. (2014) compared visual and
auditory stimuli in CBM-I, but found no significant differences. In terms
of stimulus orientation as a moderator of AB change, Heeren et al.
(2015) found side-side training was significantly more effective than
top-bottom, and conversely Hakamata et al. (2010) revealed top-
bottom training to be significantly more effective than side-side.
Out of six meta-analyses, three found number of sessions to be a
significant moderator of CBM. Hakamata et al. (2010) and Menne-
Lothmann et al. (2014) found that more sessions of ABM produced
greater change in AB. However, Cristea et al. (2016) found a single
session of CBM was significantly more effective at inducing a change in
bias than multiple sessions.
In addition, the following moderators were identified in a single
meta-analysis. ABM changed biases more when the protocol had more
distinct (rather than repeated) trials, and a larger total number of trials
(Heeren et al., 2015). CBM-I generated larger positive bias change when
participants received a lower ratio of active to total training items, were
given feedback about their performance and when imagery instructions
were used (Menne-Lothmann et al., 2014). Additionally, the AS para-
digm was more successful at inducing a positive interpretational style
compared with other CBM-I protocols, however when measuring
change in positive bias (pre-post), the other paradigms were more ef-
fective.
3.2.5. Training location
Six studies compared delivery location as a moderator on CBM ef-
ficacy. Two of the six (Linetzky et al., 2015; Mogoaşe et al., 2014) found
ABM to be significantly more potent on AB change when it was deliv-
ered in the laboratory compared with home.
3.2.6. Bias and symptoms at baseline
Two studies explored the effect of pre-existing AB on the efficacy of
ABM. Mogoaşe et al. (2014) found that a stronger baseline AB was
associated with more AB change following training. However, Heeren
et al. (2015) did not find any effect.
3.3. Do CBM protocols change symptoms?
3.3.1. Anxiety symptomatology
Of the 10 studies that measured changes in anxiety symptoms, 8
found significant results, which ranged from small to large effects (ES
= 0.13–0.74). Of the two meta-analyses that did not find an impact,
one was solely with children (Cristea et al., 2015b) and the other
(Kampmann et al., 2016) included forms of CBM that were not included
in other meta-analyses (e.g. Approach-Avoidance task ‘AAT’2). Ad-
ditionally, neither of these two meta-analyses assessed change in bias.
Hence, overall, there appears to be good evidence to support the effi-
cacy of CBM for anxiety symptoms in adults.
3.3.2. Depressive symptomatology
Depressive symptomatology was assessed as an outcome in seven
meta-analyses. Only three analyses found CBM to significantly modify
symptoms (ES = 0.33–0.85). Of those seven meta-analyses, only one
(Mogoaşe et al., 2014) had an inclusion criteria that studies must have
assessed depressive symptoms as a primary outcome. However, Mo-
goase et al. found improvement in symptoms that was greater for the
control than the ABM group. Of the remaining six meta-analyses,
Hakamata et al. (2010) yielded a large effect of ABM. While Cristea
et al. (2015a) revealed a small, but significant effect of CBM on de-
pressive symptoms; it was not significant following publication bias
analyses. These results suggest that any effect of CBM on depressive
symptoms is less robust and smaller than the effect of CBM on anxiety.
Menne-Lothmann et al. (2014) investigated the effect of CBM-I on
negative mood. They found CBM-I significantly changed participants’
mood in line with direction of training. That is, benign training im-
proved mood whereas negatively valenced training led to an increase in
negative mood.
3.3.3. Appetitive stimuli (eating disorders, smoking and alcohol use)
There were three comparisons from two meta-analyses that ex-
amined the effect of ABM on appetitive stimuli. Both meta-analyses
found evidence of small effects of ABM on reducing appetitive beha-
viour (ES = 0.003–0.36), however for Beard et al. (2012) the effect was
only significant when the neutral training was compared to training
towards appetitive stimuli. In Beard et al. (2012) ABM significantly
improved vulnerability to an appetitive stressor, whereas in Mogoaşe
et al. (2014) ABM reduced substance use symptomatology. The Cristea
et al. (2016) meta-analysis combined the effects of ABM, AAT and
2
The Approach-Avoidance task (AAT) is another form of CBM that is used in the ad-
diction literature, but not commonly applied to other psychopathologies. This involves re-
training participants to make avoidance movements (pushing a joystick away) when
presented with alcohol-related pictures, which contrasts to their usual tendency to ap-
proach (pulling a joystick towards) such stimuli (Weirs et al., 2013).

177
E.B. Jones, L. Sharpe Journal of Affective Disorders 223 (2017) 175–183

response inhibition tasks and found no overall impact of CBM on ad- 3.5. Who does CBM work best for, and under what conditions?
diction or craving outcomes. Hence, there is weak evidence for efficacy
of ABM for appetitive behaviours. 3.5.1. Age and gender
Of the six meta-analyses that assessed age as a moderator of CBM
3.3.4. Follow-up outcomes efficacy, only two revealed a significant effect of CBM on symptoms.
Follow-up measures for ABM ranged from 1 to 6 months and were Mogoaşe et al. (2014) and Price et al. (2016) indicated similar results;
reported in four analyses (Heeren et al., 2015; Kampmann et al., 2016; that ABM was more effective for anxiety reduction in younger partici-
Linetzky et al., 2015; Mogoașe et al., 2014). None of the analyses found pants. However, the majority of included studies were with adults. Si-
a lasting effect of ABM on reduction of emotional outcomes (mainly milarly, three meta-analyses investigated gender as a moderator. Only
anxiety). However, despite no significant post treatment outcomes of Menne-Lothmann et al. (2014) found a moderation effect and only for
CBM in the Cristea et al. (2016) analysis, the authors reported a small, negative mood. Hence, there is only weak evidence to suggest that
significant effect on follow up addiction measures (mean = 6 months). demographic factors might moderate the efficacy of CBM protocols.
Therefore, while there is currently an absence of evidence for long term
change in emotional outcomes, CBM may produce longer term change 3.5.2. Sample and psychopathology type
in addiction outcomes. Four studies compared the moderating effects of specific psycho-
pathology on CBM efficacy. Only Mogoaşe et al. (2014) found a dif-
3.4. Mechanism for change ference across anxiety presentations, with ABM being most effective for
generalized anxiety (ES = 0.61).
CBM is hypothesized to improve symptoms by changing the biases Ten comparisons from eight analyses compared clinical, subclinical
that are targeted. Hence, it is of interest to determine whether the and healthy samples to determine whether the efficacy of CBM differed
change in bias is associated with changes in outcomes. Three meta- across anxiety severity. Cristea et al. (2015a) found that CBM was more
analyses investigated this question, and 2/3 found a significant re- effective for symptom control in subclinical rather than clinical samples
lationship. In the Mogoaşe et al. (2014) meta-analysis, AB change for social anxiety. None of the other meta-analyses found anxiety se-
correlated positively and significantly with symptom change, whereas verity to be a moderator.
in Menne-Lothmann et al. (2014) associations were observed between
induced interpretation bias and mood. Hakamata et al. (2010) did not
3.5.3. Bias and symptoms at baseline
find a significant relationship, but revealed a trend for a positive re-
Three studies looked at pre-existing bias as a moderator of CBM
lationship between change in AB and anxiety symptoms.
efficacy. Price et al. (2016) was the only one to find participants’
Four meta-analyses attempted mediation analyses. However, Price
baseline AB significantly moderated the efficacy of ABM. Higher base-
et al. (2016) was the only analysis to find evidence for this relationship.
line AB predicted better outcomes in ABM, but only for participants
The authors could not confirm mediation with the full sample, but
who were trained under optimal conditions (clinician rated anxiety and
showed mediation was present in subgroups where ABM was particu-
laboratory trained). In contrast, Heeren et al. (2014) found participants
larly effective. Therefore, although there is relatively strong evidence
with lower State Trait Anxiety Inventory (STAI) trait scores benefitted
that the more that CBM modifies cognitive processes, the more that
more from ABM in reducing stress vulnerability.
subsequent emotional outcomes change, further evidence is required to
substantiate the mechanistic ties between changes in targeted bias and
symptom reduction. 3.5.4. Symptom control versus vulnerability to stress
Broadly in the literature, CBM has been applied in two different
Table 2 ways: to reduce vulnerability to a future stressor; or to improve psy-
Summary of main findings. chological symptoms in participants. Symptom control and stress vul-
nerability measures were compared in six of the analyses. Four meta-
ABM
analyses investigated the efficacy of ABM on vulnerability to stress or
Finding Weight of Evidence anxiety following a stressor. All studies that examined this revealed
evidence of a protective impact of ABM, with small to moderate effects
Protocols do change attention biases in adults 10/10 meta-analyses (ES = 0.37 – 0.77). However, within Beard et al. (2012), ABM was
Protocols improve stress vulnerability 5/5 meta-analyses effective only in the neutral (training towards neutral words or faces) vs
Protocols improve anxiety in adults 8/9 meta-analyses
No evidence that ABM is effective in children 1/1 meta-analysis
disorder relevant (training towards threat or appetitive stimuli) con-
Change in bias correlates with change in 1/2 meta-analyses dition. Furthermore, Hallion and Ruscio (2011) explored a combination
symptoms of ABM and CBM-I, and also found a small buffering effect on stress
Evidence of full mediation not supported 4/4 meta-analyses vulnerability. In contrast, only one study looked at the efficacy of CBM-I
Effects are larger for stress vulnerability than self- ES ranges (Anxiety: 0.16 −
on vulnerability to stress (Menne-Lothmann et al., 2014), and did not
report anxiety symptoms 0.41;
Stress vulnerability: find a significant effect. Hence, there seems to be good evidence, at
0.37–0.77) least where disorder-specific vs neutral stimuli are used that ABM is
efficacious in improving emotional outcomes to a stressor.
CBM-I Four studies analysed the effect of ABM on overall symptom control,
Finding Weight of Evidence
with all revealing small to moderate effects (ES = 0.16–0.41).
However, in Beard et al. (2012), training was effective only in the
Protocols do change interpretive biases in adults 2/2 meta-analyses neutral (training towards neutral words or faces) vs control (no active
Protocols do change interpretive biases in 1/1 meta-analysis training) condition. Menne-Lothmann et al. (2014) was again the only
children, but not symptom reduction
study to explore the effect of CBM-I alone, and they found small im-
Protocols do not improve stress vulnerability 1/1 meta-analyses
Protocols improve anxiety in adults 2/2 meta-analyses provements in symptom reduction following benign training, and small
Protocols modify negative mood states in adults 1/1 meta-analyses increases in symptomatology following negative training. Two studies
Change in bias correlates with change in 1/1 meta-analyses examined the combined effect of ABM and CBM-I on symptom reduc-
symptoms
tion, with one study (Hallion and Ruscio, 2011) again finding a small
Evidence of full mediation is not supported 2/2 meta-analyses
effect. Cristea et al.'s. (2015b) child-focused meta-analysis failed to find
an effect of ABM on symptom control.

178
E.B. Jones, L. Sharpe
3.5.5. Relative efficacy of CBM approaches
Three meta-analyses compared ABM to CBM-I, and one study
(Cristea et al., 2015a) found a significant difference on anxiety, general
anxiety and depressive symptoms favouring CBM-I. Neither Hallion and
Ruscio (2011) nor Cristea et al. (2015b) found differences in efficacy
between CBM-I or ABM.
It appears overall that CBM-I may have more power as a paradigm
in symptom reduction compared with ABM, whereas ABM may be more
powerful as a buffer to stress vulnerability. However, moderator ana-
lyses have not been applied.
3.5.6. Protocol parameters
Stimulus modality was tested as a moderator of ABM efficacy in
three studies. The two studies (Beard et al., 2012; Hakamata et al.,
2010) comparing verbal targets (words) to pictorial targets, found that
words were more efficacious on symptoms. Menne-Lothmann et al.
(2014) compared the visual and auditory stimuli in CBM-I, but found no
significant differences.
Four studies investigated whether the orientation of the training
stimuli moderated ABM potency. Two (Beard et al., 2012; Hakamata
et al., 2010) found top-bottom rather than side-side orientation led to a
larger effect size for symptom control. In contrast, Heeren et al. (2015)
found side-side presentation to be more effective for stress vulner-
ability, and no difference for symptom reduction. Of note is that in
Heeren et al. (2015), 14/15 studies utilised solely face (pictorial) sti-
muli, whereas in Beard et al. and Hakamata et al. the ratio was 24/41
and 7/12 respectively. Therefore, it may be that words are better pre-
sented in top-down presentations, but pictures should be presented
side-by-side.
Training paradigm was examined in two analyses, with both finding
significant effects. Menne-Lothmann et al. (2014) found that CBM-I
paradigms other than AS were more effective in reducing negative
mood. Mogoaşe et al. (2014) reported the dot-probe as more powerful
in reducing anxiety than the visual cueing ABM task. Neither meta-
analysis found significant effects of paradigm on stressor vulnerability.
Training stimuli (i.e. neutral vs positive) was compared as a mod-
erator in two studies, but neither found a significant difference (Heeren
et al., 2015; Mogoașe et al., 2014).
In terms of dose, number and timing of training trials did not
moderate outcomes in the two meta-analyses that examined this
question (Heeren et al., 2015; Mogoașe et al., 2014). Similarly, time
between sessions was not a significant moderator (Cristea et al., 2015b;
Mogoașe et al., 2014).
Three out of 9 meta-analyses found number of sessions to be a
moderator. Beard et al. (2012) and Menne-Lothmann et al. (2014)
found that more sessions increased the potency of ABM and CBM-I on
symptom control. In contrast, Cristea et al. (2015a) found that fewer
sessions of CBM lead to a larger effect on symptom reduction.
A number of other stimulus characteristics have been investigated in
single meta-analyses and found to moderate efficacy of CBM. According
to these analyses, ABM was more effective as a stressor buffer when
there was a larger number of distinct (rather than repeated trials)
(Heeren et al., 2015). Further, ABM outcomes were more pronounced
for stress vulnerability and symptom reduction when primary rather
than secondary outcome measures were used (Mogoaşe et al., 2014).
CBM-I was more effective at reducing stress vulnerability if implicit
(rather than explicit instructions) were given, and more effective at
reducing negative mood if participants were given imagery instructions
and did not generate words (for paradigms other than ambiguous sce-
narios).
3.5.7. Outcome measure
Three of five meta-analyses found that CBM was more effective on
some outcomes than others. Hakamata et al. (2010) found that trait
anxiety yielded a larger effect of ABM than state anxiety. In Price et al.
(2016), ABM appeared to be more beneficial according to clinician
179
Journal of Affective Disorders 223 (2017) 175–183
rated rather than self-reported anxiety measures. The same pattern was
observed in Linetzky et al. (2015) in a clinical population; however it
was not formally tested as a moderator. The effect sizes in the Mogoaşe
et al. (2014) analysis also suggest that ABM is more effective following
clinician rated anxiety, although the authors did not report any sig-
nificant differences. Cristea et al. (2015b) compared scales that were
validated to all of their included measures, but found no differences.
3.5.8. Training location
Out of the eight studies that included location as a moderator, five
found significant effects. All five studies found evidence to support a
larger effect of CBM in the laboratory compared with remote admin-
istration (Cristea et al. (2015a), Heeren et al. (2015) Linetzky et al.
(2015); Kampmann et al. (2016); Price et al. (2016). The remaining
meta-analyses found no differences (Cristea et al., 2016, 2015b;
Mogoașe et al., 2014) There appears to be good evidence supporting the
laboratory as a preferable location for CBM training compared with the
home.
3.5.9. Compensation
Two studies investigated the effects of compensation on symptom
control, but only Cristea et al. (2015a) reported a significant effect. For
all anxiety, general anxiety and social anxiety measures, participants
showed a greater reduction in symptomatology if they were compen-
sated for their time.
3.6. Methodological issues
Publication year was a significant moderator in two studies (Cristea
et al., 2015a; Heeren et al., 2015), but not in the other two where it was
included (Beard et al., 2012; Cristea et al., 2015b). Quality score was
significant in the same two studies, but again, not in Cristea et al.
(2015b). Journal Impact Factor (JIF) was significant only in Heeren
et al. (2015), but not in Cristea et al. (2015a), (2015b). These results
indicate that although ABM and CBM-I appear to have robust effects,
poorer quality and earlier studies published in journals with a higher
JIF found larger effects. Therefore, the effect sizes for CBM may not be
as large as the original meta-analyses estimated.
Table 3
Recommendations for parameters to use for CBM.
ABM
ABM appears best suited to use to reduce stress vulnerability in adults
Lower trait anxiety people in particular seem to benefit from ABM. Inconsistent
evidence for other demographic/ baseline characteristics
ABM is most effective when delivered in the laboratory
Pictures should be presented left-right
Words should be presented top-down
Based on single meta-analyses, the following ABM parameters should be adopted:
Use of the dot-probe rather than spatial cueing task
Implicit instructions
More distinct rather than repeated trials
CBM-I
CBM-I appears best suited to use to reduce anxiety symptoms in the short-term in
adults
Women in particular appear to benefit from CBM-I. Inconsistent evidence for other
demographic/ baseline characteristics
CBM-I is most effective when delivered in the laboratory
Based on 1 meta-analysis, the following CBM-I parameters should be adopted:
Inclusion of filler trials
Instructions to imagine the scenarios
Requirement to generate words for AS task, but no word generation for other
tasks
Provision of feedback
AS: ambiguous scenarios
E.B. Jones, L. Sharpe
4. Discussion
4.1. Summary of findings
In reviewing the twelve existing meta-analyses on CBM, our ques-
tion was twofold: Does CBM work, and if so, under what conditions is it
most effective? ABM changed attentional biases in adults in 8/8 meta-
analyses, indicating strong evidence for its effect. However ABM did not
successfully modify attentional biases in the single meta-analysis to
include only child studies. Similarly, CBM-I modified interpretive biases
in adults in 2/2 meta-analyses, but also successfully changed inter-
pretive biases in the child only sample. Therefore, ABM and CBM-I
protocols consistently modified targeted biases in adults. ABM as a sole
intervention improved anxiety symptoms for adults in all six meta-
analyses where it was measured, and likewise, the two meta-analyses
that combined ABM and CBM-I analyses revealed improved anxiety in
adults. However, the results were not as compelling for depressive
symptomatology. Taking into account publication bias, only 2/7 ana-
lysis revealed CBM to be effective at reducing depressive symptoms.
ABM was associated with small effects on appetitive symptoms in two
analyses. Across six of the meta-analyses, CBM was used in two dif-
ferent applications: symptom reduction and stress vulnerability. It ap-
pears that although ABM did significantly reduce symptomatology, its
most effective application is ameliorating vulnerability to a stressor.
CBM-I on the other hand appears to work better for symptom reduction.
From the current results, it can be concluded that there is good evidence
for CBM's efficacy in changing targeted biases, and improving anxiety
and stress vulnerability.
There are several recommendations for CBM parameters that
emerged from our review of the literature, and most were specific to
either ABM or CBM-I protocols. One finding that was consistent across
both protocols however was that CBM is most effective when delivered
in the laboratory. For ABM, individuals with lower trait anxiety scores
appeared to benefit more from the training when faced with a stressor.
There was little evidence to support the moderating effect of any other
demographic or baseline characteristics, and where evidence was
available; the outcomes were inconsistent across meta-analyses. There
were mixed findings when it came to the effect of stimuli (word vs
picture) and orientation (top-down vs side-by-side presentation).
However, careful attention to the relationship between these two re-
vealed that in cases where word stimuli were used, training was more
effective with a top-down orientation. The opposite was true for pic-
torial stimuli, where a side-side orientation proved to be more effective.
Further, the dot-probe appears to yield larger effects on anxiety re-
duction than the spatial cueing task. More distinct rather than repeated
stimulus pairs also seem to generate larger effects of ABM on AB change
and as a stress buffer. Lastly, implicit instructions seem to be more ef-
fective in inducing a prophylactic effect on stress vulnerability, rather
than informing participants of the possible therapeutic nature of the
training. However, before conclusions are made regarding the issue of
expectancy effects in CBM, it seems important to analyze the role of
control conditions and treatment expectancy effects to determine the
relative benefit of CBM over and above placebo response.
CBM-I on the other hand seems to particularly benefit females ra-
ther than males, however there was again little and inconsistent evi-
dence to suggest other demographic or baseline characteristics differ-
entially affect the training. The following parameters appear to increase
the efficacy of CBM-I: inclusion of filler trials, instructions to imagine
the scenarios, and feedback about response accuracy. When partici-
pants were required to generate the meaning of words and situations,
this increased the efficacy of the ambiguous scenarios paradigm.
However, using other paradigms such as the WSAP or homograph task,
the effectiveness of CBM-I was reduced when word generation was
employed.
180
Journal of Affective Disorders 223 (2017) 175–183
4.2. ABM compared to CBM-I
Previous research has demonstrated transfer effects between ABM
and CBM-I (e.g. Amir et al., 2010; White et al., 2011). This suggests
interpretive and attentional biases are not independent cognitive me-
chanisms, but that there is a shared underlying mechanism between
these selective processing systems. Indeed, these findings support cog-
nitive theories of anxiety and depression (e.g. Beck, 1976; Clark and
Wells, 1995; Mathews and MacLeod, 2005) in their proposition that
while distinct, cognitive mechanisms are interactive in nature. Simi-
larly, results from the current review suggest ABM and CBM-I are un-
ique from one another and may have distinct applications. ABM yielded
larger effects as a buffer to stressor vulnerability compared with
symptom reduction measures, whereas CBM-I did not appear to affect
emotional reactivity. However, Menne-Lothmann et al. (2014) propose
the reason for their finding may have been that the stimuli were not
ambiguously stressful, but rather universally distressing. Conversely,
CBM-I significantly outperformed ABM on post treatment measures of
anxiety and depressive symptomatology in one analysis, which may
indicate its suitability as a method of symptom reduction over and
above ABM. It has been suggested that CBM-I procedures in particular
may prime participants to respond in either a negative or positive
manner and allow them to respond more quickly to targets that match
the valence of the training without altering the interpretation style
(MacLeod and Mathews, 2012). However, studies (e.g. Hoppitt et al.,
2010) have compared passive exposure to valenced stimuli with active
CBM-I training, and found only the active training to affect symptoms.
Further, changes following CBM-I seem to last longer than priming ef-
fects (MacLeod and Mathews, 2012). Cristea et al. (2015a) suggest
CBM-I is more vulnerable to demand characteristics given the partici-
pants are more aware of the direction of training, which may explain its
increased impact on biases. Although this may be true, MacLeod and
Mathews (2012) argue that despite the direction of the training po-
tentially being evident, participants are unlikely to accurately gauge the
expected responses on bias tests and outcome measures. Also, partici-
pants are often unable to differentiate between control and outcome
variables, and do not reliably report expectancies (MacLeod and
Mathews, 2012). It appears therefore, that CBM-I protocols may cur-
rently be more effective at evoking changes on targeted biases than
ABM. However, ABM paradigms are still successfully inducing bias
change, and in fact appear more powerful than CBM-I at reducing stress
vulnerability. However, it is important to note that these conclusions
should be considered to be hypothesis-generating, in that, there have
not been direct comparisons between ABM and CBM-I to support them
within an experiment. However, more research should be conducted to
confirm these patterns in order to ensure that CBM protocols can be
used in situations where they are most likely to efficacious.
4.3. Clinical utility
ABM and CBM-I have been employed to modify symptoms from a
range of psychopathologies, and have also been used to improve
symptomatology in analogue or healthy samples. Outcomes on anxiety
measures were robust, and encourage future use of ABM and CBM-I to
reduce anxiety symptomatology. However, none of the analyses re-
vealed a lasting effect of CBM on reduction of anxiety, or any other
emotional outcomes. While no significant post-treatment outcomes
were found in the Cristea et al. (2016) analysis, the authors reported a
small, significant effect of CBM on follow-up addiction measures (mean
= 6 months). Therefore, there is currently only evidence that CBM may
produce longer term changes in addiction outcomes. Despite the
breadth of previous applications, the current review found inconsistent
evidence regarding the use of CBM in reduction of depressive symp-
toms. It is notable that only one meta-analysis specifically assessed
symptoms associated with depression as a primary outcome, rather
than a secondary, complimentary measure in an anxiety study. When
E.B. Jones, L. Sharpe
other analyses did include studies targeting depressive symptoms as a
primary outcome, these were combined with secondary measures. In
their moderator analysis, Mogoaşe et al. (2014) found primary outcome
measures generated larger effects than secondary for both symptom
reduction and stress vulnerability, demonstrating the importance of
distinguishing primary from secondary measures in future meta-ana-
lyses. Also, in cases where depression and anxiety were compared in
moderation analyses of pathology type, there were no significant dif-
ferences. Therefore, it is perhaps premature to conclude that ABM is not
effective for depressive symptoms. Although, it is worth noting that
Mogoaşe et al. (2014) did analyze depressive symptomatology as a
primary measure only, and reported the control group improving more
in mood than the experimental ABM group. Even though it is difficult to
ascertain the clinical utility of CBM more generally on symptoms of
depression, it is apparent from the results that protocols aimed to
change anxiety do not have ensuing effects on depressive symptoma-
tology.
CBM for younger people is a more recent area of research, for which
the efficacy is unclear. Four meta-analyses included in this review
(Beard et al., 2012; Linetzky et al., 2015; Mogoaşe et al., 2014; Price
et al., 2016) incorporated some data from child samples in their ana-
lysis, but the percentage of adult studies was markedly higher than
child studies (5–10%). Mogoaşe et al. (2014) found in their anxiety
subsample that not only was ABM more effective at modifying biases in
younger relative to older people, but also that younger people bene-
fitted more in anxiety reduction. Similarly, Price et al. (2016) revealed
that ABM led to reductions in social anxiety only when participants
were less than 37 years. Interestingly, Pergamin‐Hight et al. (2016)
found in their RCT that older but not younger children's anxiety sig-
nificantly reduced following ABM. Considering the aforementioned
findings, it is plausible that the relationship between age and ABM
potency may assume an inverted U-shape, such that it is most effective
for adolescents and young adults (Pergamin‐Hight et al., 2016). In
contrast to this, however, Cristea et al. (2015b) included only child
studies in their meta-analysis (mean age = 12.71) and did not find
ABM to effectively change biases, or lead to improvements in symp-
toms. The majority of ABM studies in Cristea et al. (2015b) utilised the
dot-probe task, with pictorial stimuli and short presentation times.
Dudeney, Sharpe and Hunt (2015) found these parameters to be sub-
optimal to detect attentional biases in children in their meta-analysis.
Further, the authors indicate that children do exhibit an attentional bias
towards threatening stimuli, albeit to a smaller degree than that found
in adults. Cristea et al. (2015b) did report CBM-I to effectively change
interpretive biases in young people, although the effects did not
translate to significant effects on outcome measurements. The current
data and analysis of the CBM child literature are currently insufficient
to support the use of CBM for children.
Despite promising results of ABM and CBM-I in the current review,
the published effect sizes are only moderate. Cognitive Behavioural
Therapy (CBT) has the most robust evidence base for the treatment of
anxiety and depression. A review of meta-analyses on CBT revealed it to
have a large grand mean effect (ES = 0.95) on depression and a range
of anxiety disorders compared to placebo, waitlist, or no-treatment
controls (Butler et al., 2006). Further, computerized CBT also appears
to have large effects (ES = 0.88) on depression and anxiety (Andrews
et al., 2010). CBM is a markedly lower-intensity intervention than CBT,
which likely explains some of the difference in power. However, there is
scope to make CBM paradigms more powerful. Results from the current
analysis showed CBM was most effective when delivered in the la-
boratory rather than at home. Given participants report CBM tasks to be
repetitive and boring (Brosan et al., 2011), it is possible that the con-
trolled laboratory testing encourages stricter adherence to instructions,
such as sustained attention to the screen and effort to respond accu-
rately. Although this somewhat restricts the clinical application of CBM,
results from computerized CBT research (e.g. Andrews et al., 2010)
indicate that when cognitive therapies are delivered via technology
181
Journal of Affective Disorders 223 (2017) 175–183
they can be effective in reducing depression and anxiety symptoms. As
such, future research could aim to increase the acceptability of CBM
tasks by making them more engaging in order to capture attention and
focus of individuals at home.
4.4. Variations between meta-analyses
A large variation was observed in the presence and extent of sig-
nificant effects in outcome measures and moderator variables across the
twelve meta-analyses. These variations arose largely from discrepancies
in the inclusion and exclusion criteria. Cristea et al. (2015b) was the
only meta-analyses not to find an effect of ABM on bias change or any
outcome measures. As previously discussed, it appears these results
derive largely from the nature of child only samples. Despite examining
the effects of CBM on outcomes measures, Cristea et al. (2015a) and
Kampmann et al. (2016) did not include bias change as a measure. It is
unclear why the authors excluded cognitive bias change as an im-
portant indicator of CBM efficacy, when the literature indicates a
change in bias is necessary for a reliable change in symptoms (e.g.
Clarke, Notebaert and MacLeod, 2014). Of the eight analyses that
measured ABM efficacy on bias change in adults, there was a large
range in estimated effects from small to large effects. Hakamata et al.
(2010) found the largest effect size for AB change compared with other
analyses. As indicated by Heeren et al. (2015), larger AB scores were
associated with earlier publications. Given Hakamata et al. (2010) is
the earliest meta-analysis of CBM, it included the fewest studies, and
was most likely to include the studies that were of poorer quality, it is
perhaps unsurprising that this meta-analysis reports a somewhat in-
flated effect size. The effect sizes for CBM-I analyses also varied, ran-
ging from moderate to large effects. Given the variation in effect sizes
on bias change, the median effect size (derived from published values in
earlier analyses) of 0.61 would seem to be a good estimate of the true
efficacy of ABM, and the median effect size of 0.65 would appear to be a
good estimate for CBM-I.
Of the ten analyses exploring anxiety symptom reduction in adults,
Cristea et al. (2015a) found the weakest significant effect. These results
are not surprising, as the authors excluded outcomes associated with
stressor vulnerability due to variability among the studies (Cristea
et al., 2015a). Given that the other meta-analyses investigating ABM
found larger effects for stressor vulnerability than symptom reduction,
this omission would have reduced the effect sizes of ABM. Also, Cristea
et al. (2015a) was the only analysis to include studies that did not
necessarily focus on clinical symptoms or distress as primary outcomes.
For example, Sharpe et al. (2012) implemented an ABM protocol with
the aim to reduce biases towards pain. Anxiety was measured as a
secondary outcome. Hence, the inclusion of studies such as this likely
underestimates the efficacy of ABM. Of all the analyses, Cristea et al.
(2015a) appeared to use the most stringent criteria for their included
studies to ensure they would only find an effect if it was valid and
reliable. Given this, the fact that the authors still found an effect of
training on anxiety reduction demonstrates that CBM protocols do ef-
fect change. Notably, a recent re-analysis of the Cristea et al. (2015a)
meta-analysis (Grafton et al., 2017) was conducted where the success of
bias modification was used as a moderator. Results confirmed that for
those studies where biases had been successfully modified, medium
effect sizes were observed (g=0.60). In contrast, where no modification
had occurred in response to CBM, Grafton et al. (2017) found that there
was no impact on emotional vulnerability. Cristea et al. (2017) subse-
quently re-analysed studies through meta-regression using anxiety as
the outcome measure, rather than emotional vulnerability. Their results
confirmed a smaller (but significant) effect size for those studies where
biases were changed (d = 0.38), and confirmed that change in bias did
predict outcome in univariate analyses (β = 0.42, p = 0.032). How-
ever, in multivariate regression, bias change no longer predicted out-
come (p > 0.05), suggesting that other variables could account for the
association. However, without additional information on the analyses,
E.B. Jones, L. Sharpe
the results are difficult to interpret.
Kampmann et al. (2016) was the only adult study to report no effect
of CBM on anxiety reduction in adults. There are a couple of reasons
this may have occurred. Firstly, the authors included data from CBM
protocols that are not well validated in the anxiety literature (e.g. AAT).
Secondly, in 6/13 included studies the location of CBM was at home,
which as previously discussed is sub-optimal. When Kampmann et al.
(2016) computed the effect of CBM administered in the laboratory they
did find a small, significant result. Furthermore, 12/13 studies relied on
the Liebowitz Social Anxiety Scale (LSAS) as the measure of anxiety. In
their meta-analysis, Price et al. (2016) demonstrate that ABM was only
more beneficial than the control according to the LSAS when partici-
pants were trained in the lab. Whereas with clinician rated anxiety,
CBM was effective at anxiety reduction regardless of location.
Variations in inclusion and exclusion criteria also altered the out-
comes of moderators. For example, Heeren et al. (2015) found side-side
training was significantly more effective than top-bottom, whereas
Hakamata et al. (2010) and Beard et al. (2012) found top-bottom
training was more effective than side-side. These opposing results could
be because Heeren et al. (2015) included a large number of studies
(93%) with pictorial training stimuli whereas Beard et al. and Haka-
mata et al., where only 58% of studies incorporated pictorial stimuli for
training. Heeren et al. (2015) suggest that it is more ecologically valid
to process faces horizontally rather than vertically. Hence, ABM studies
using words may benefit from top-down orientation, whereas those
using pictorial stimuli may increase training effectiveness through left-
right orientation. As further evidence, Hakamata et al. and Beard et al.
found that in those conditions where top-bottom orientation was more
effective, words were also more effective than pictures as training sti-
muli. However, there is large variability in a range of parameters in-
cluding presentation time, stimuli, stimuli orientation, which may well
have an impact on the relative efficacy of different paradigms, but are
rarely directly compared (Lopes et al., 2015). The disparities in out-
comes between the meta-analyses indicate why the current review is
useful in forming recommendations for future CBM research.
4.5. CBM as a process
There has been a tendency for authors to publish null findings as
evidence against the efficacy of CBM, even if the procedure used to
induce a change in cognitive bias did not induce the desired change.
However, if a paradigm does not reliably alter either attentional or
interpretive biases, then it cannot be called ABM or CBM-I. In this way,
it has been argued that CBM should be conceptualized as a process,
rather than a procedure (MacLeod and Grafton, 2016). Previous re-
search (Clarke et al., 2014; Grafton et al., 2017) has indicated that
when tasks are successful in changing attention, they reliably lead to
changes on outcomes. It is therefore self-evident that a study failing
firstly to modify attention should also fail to produce emotional change.
As previously mentioned, all meta-analyses except Cristea et al. (2015a)
and Kampmann et al. (2016) included bias change as an outcome
measure. All three analyses found an effect of CBM-I on bias change,
and 8/9 studies detected bias change following ABM. The one analysis
that did not report an effect of ABM on bias change was a child only
sample (Cristea et al., 2015b). These results then support the argument
that producing change in cognitive biases is necessary for the sub-
sequent change in emotional outcomes. In fact, 2/3 meta-analyses in-
dicated that greater levels of bias change correlated with greater
symptom improvement. Despite this result, none of the four meta-
analyses that conducted mediation analyses found it to be present in the
full sample. It is important here to realize that achieving mediation in
meta-analyses is difficult given the substantial heterogeneity in out-
comes. However, in Price et al. (2016) mediation was present in sub-
groups (e.g. younger participants, laboratory training, assessed by
clinicians) where ABM was particularly efficacious. Furthermore,
mediation has been shown at the individual study level (e.g. Amir et al.,
182
Journal of Affective Disorders 223 (2017) 175–183
2010; See et al., 2009), although many studies have failed to find
mediation. For a fair test of mediation, an individual study firstly needs
to have reliably elicited a change in targeted biases, and secondly
modified outcome measures, but then show that the main effect of
treatment was mediated by the change in bias. Future research should
aim to meet these standards, so that there is more clarity about the
effectiveness of CBM in cases where the process was actually evoked.
This can then inform the development of more powerful paradigms to
produce cognitive bias modification.
4.6. Limitations
Although the current review aimed to elucidate meta-analytic
findings of CBM, it is not without its own limitations. Firstly, the in-
cluded meta-analyses that combined results for ABM and CBM-I (Cristea
et al., 2015a, 2015b; Hallion and Ruscio, 2011) did not allow for a
breakdown of effect sizes for each type of protocol (i.e. ABM or CBM-I)
on outcome measures. This means that it was difficult to ascertain what
the individual effects of each protocol were, and therefore which pro-
tocol was most effective for each outcome. Further, the blending of
ABM and CBM-I reporting may have led to one modification type
compensating for a lack of effect on behalf of the other modification
type, or may have reduced the effect size on outcomes. Indeed, Hallion
and Ruscio (2011) yielded the weakest effect for stressor vulnerability
when compared to ABM only analyses, supporting the idea that ABM is
more effective as a stress buffer than CBM-I. The same effects may have
also occurred in Cristea et al. (2016) or Kampmann et al. (2016) fol-
lowing the combined analyses of ABM with AAT and response inhibi-
tion tasks or AAT and Attention Interpretation Modification (AIM) re-
spectively. While one method to overcome this limitation would be to
conduct a grand meta-analysis, we felt that the plethora of quantitative
analyses already available for CBM indicated that a qualitative narra-
tive review at this point may elucidate more important patterns to
generate hypotheses for future research.
Another related limitation was the mere presence of one CBM-I only
meta-analysis, amongst five ABM only meta-analyses. It is therefore
challenging to determine the efficacy of CBM-I in multiple contexts,
especially when the aforementioned combined analyses do not specify
individual effect sizes. For example, none of the analyses investigated
the efficacy of CBM-I specifically on anxiety reduction, and so results
demonstrating this finding represent combined effects of ABM and
CBM-I, and are therefore preliminary. In this way, interpretations about
the nature and recommendations for CBM-I training are limited in
comparison to ABM, and inferences based on these results in the current
review should be made with this in mind.
The nature of the meta-analytic process requires averaging effect
sizes across different studies. This is useful for identifying overarching
patterns, but restricts access to a wider depth of information. For ex-
ample, the individual studies in each meta-analysis that generated
particularly large or small effects were excluded, and therefore do not
inform conclusions made in this review. The literature may actually be
able to establish more clarity regarding the effectiveness of specific
CBM parameters by comparing protocols directly within the same
study. For example, Sharpe et al. (2015) compared words and faces
(using a top-down presentation), and found words to be more effica-
cious at modifying attentional biases. Similarly, Todd et al. (2016) in-
vestigated specific word types in pain and found affective training to be
most efficacious, and Bowler et al. (2016) found ABM training at
500 ms to be more effective than 1250 ms. These within subjects results
are arguably more rigorous than meta-analysis, and can guide research
practices for quite specific applications. Future research should con-
tinue to focus on comparing different parameters within individual
studies, as well as other patterns that emerge from the data, such as the
differential effects of CBM-I versus ABM on anxiety reduction and stress
vulnerability.
E.B. Jones, L. Sharpe
5. Conclusion
Results from the current review indicate CBM consistently modifies
targeted biases in adults, but not children. In terms of symptomatic out-
comes, ABM appears to be more effective as a buffer to stressor vulner-
ability, although also reduces symptomatology. On the other hand, CBM-I
does not as reliably reduce stressor vulnerability, but according to one meta-
analysis is more effective than ABM at modifying anxiety and depressive
symptoms. Follow-up results were less robust, with only one study sup-
porting the long-term efficacy of CBM for substance use. CBM appears to
have utility in improving anxiety symptoms, but the clinical utility of CBM
for depressive symptomatology is still unclear. Suggested parameters for
increasing efficacy of ABM and CBM-I are unique to each protocol, however
laboratory-based training appears to be more effective than home-based
training for both. This review proposes that the more effective CBM is at
modifying cognitive processes, the greater the potential for change on
subsequent emotional outcomes. Despite encouraging results, the effects of
existing protocols are only small to moderate. In order to impact more
change on outcomes, CBM protocols need to be more powerful and reliable
in evoking the process. CBM research in the future should always seek to
successfully modify biases before aiming to alter outcome measures, and
include mediation analyses to substantiate the mechanisms of treatment.
Some recommendations are given here to suggest ways to increase the ef-
ficacy of CBM. Future research should harness these recommendations to
adapt current CBM protocols, but should also consider how these sugges-
tions may inform the development of more reliable CBM paradigms.
Funding
This research did not receive any specific grant from funding
agencies in the public, commercial, or not-for-profit sectors.
Contributions
Authors A and B designed the review. Author A conducted the lit-
erature search, and author B was the second reviewer. Author A wrote
the first draft of the review, and author B contributed to and approved
the final manuscript.
Conflict of interest
The authors declare no conflict of interest.
References
Amir, N., Bomyea, J., Beard, C., 2010. The effect of single-session interpretation mod-
ification on attention bias in socially anxious individuals. J. Anxiety Disord. 24 (2),
178–182.
Amir, N., Taylor, C.T., 2012. Interpretation training in individuals with generalized social
anxiety disorder: a randomized controlled trial. J. Consult. Clin. Psychol. 80 (3), 497.
Andrews, G., Cuijpers, P., Craske, M.G., McEvoy, P., Titov, N., 2010. Computer therapy
for the anxiety and depressive disorders is effective, acceptable and practical health
care: a meta-analysis. PLoS One 5 (10), e13196.
Beck, A.T., 1976. Cognitive therapy and the emotional disorders. New York.
Beard, C., Sawyer, A.T., Hofmann, S.G., 2012. Efficacy of attention bias modification
using threat and appetitive stimuli: a meta-analytic review. Behav. Ther. 43 (4),
724–740.
Blackwell, S.E., Holmes, E.A., 2010. Modifying interpretation and imagination in clinical
depression: a single case series using cognitive bias modification. Appl. Cogn.
Psychol. 24 (3), 338–350.
Bowler, J.O., Bartholomew, K.J., Kellar, I., Mackintosh, B., Hoppitt, L., Bayliss, A.P.,
2016. Attentional bias modification for acute experimental pain: a randomized
controlled trial of retraining early versus later attention on pain severity, threshold
and tolerance. Eur. J. Pain 21 (1), 112–124.
Brosan, L., Hoppitt, L., Shelfer, L., Sillence, A., Mackintosh, B., 2011. Cognitive bias
modification for attention and interpretation reduces trait and state anxiety in an-
xious patients referred to an out-patient service: results from a pilot study. J. Behav.
Ther. Exp. Psychiatry 42 (3), 258–264.
Butler, A.C., Chapman, J.E., Forman, E.M., Beck, A.T., 2006. The empirical status of
cognitive-behavioral therapy: a review of meta-analyses. Clin. Psychol. Rev. 26 (1),
17–31.
183
Journal of Affective Disorders 223 (2017) 175–183
Clark, D.M., Wells, A., 1995. A cognitive model of social phobia. Social. phobia: Diagn.
Assess. Treat. 41 (68), 00022–00023.
Clarke, P.J., Notebaert, L., MacLeod, C., 2014. Absence of evidence or evidence of ab-
sence: reflecting on therapeutic implementations of attentional bias modification.
BMC Psychiatry 14 (1), 1.
Cristea, I.A., Kok, R.N., Cuijpers, P., 2016. The effectiveness of cognitive bias modification
interventions for substance addictions: a meta-analysis. PLoS One 11 (9), e0162226.
Cristea, I.A., Kok, R.N., Cuijpers, P., 2017. Invited Commentary on: Confusing procedures
with process in cognitive bias modification research.
Cristea, I.A., Kok, R.N., Cuijpers, P., 2015a. Efficacy of cognitive bias modification in-
terventions in anxiety and depression: meta-analysis. Br. J. Psychiatry 206 (1), 7–16.
Cristea, I.A., Mogoașe, C., David, D., Cuijpers, P., 2015b. Practitioner Review: cognitive
bias modification for mental health problems in children and adolescents: a me-
ta‐analysis. J. Child Psychol. Psychiatry 56 (7), 723–734.
Dandeneau, S.D., Baldwin, M.W., Baccus, J.R., Sakellaropoulo, M., Pruessner, J.C., 2007.
Cutting stress off at the pass: reducing vigilance and responsiveness to social threat by
manipulating attention. J. Personal. Soc. Psychol. 93 (4), 651.
Dudeney, J., Sharpe, L., Hunt, C., 2015. Attentional bias towards threatening stimuli in
children with anxiety: a meta-analysis. Clin. Psychol. Rev. 40, 66–75.
Fox, E., Russo, R., Bowles, R., Dutton, K., 2001. Do threatening stimuli draw or hold
visual attention in subclinical anxiety? J. Exp. Psychol.: Gen. 130 (4), 681.
Grafton, B., MacLeod, C., Rudaizky, D., Holmes, E.A., Salemink, E., Fox, E., Notebaert, L.,
2017. Confusing procedures with process when appraising the impact of cognitive
bias modification (CBM) on emotional vulnerability: A response to Cristea et al.,
(2015). Br. J. Psychiatry (in press).
Grey, S.J., Mathews, A.M., 2009. Cognitive bias modification–Priming with an ambiguous
homograph is necessary to detect an interpretation training effect. J. Behav. Ther.
Exp. Psychiatry 40 (2), 338–343.
Hakamata, Y., Lissek, S., Bar-Haim, Y., Britton, J.C., Fox, N.A., Leibenluft, E., Pine, D.S.,
2010. Attention bias modification treatment: a meta-analysis toward the establish-
ment of novel treatment for anxiety. Biol. Psychiatry 68 (11), 982–990.
Hallion, L.S., Ruscio, A.M., 2011. A meta-analysis of the effect of cognitive bias mod-
ification on anxiety and depression. Psychol. Bull. 137 (6), 940.
Heeren, A., Mogoașe, C., Philippot, P., McNally, R.J., 2015. Attention bias modification
for social anxiety: a systematic review and meta-analysis. Clin. Psychol. Rev. 40,
76–90.
Hoppitt, L., Mathews, A., Yiend, J., Mackintosh, B., 2010. Cognitive mechanisms un-
derlying the emotional effects of bias modification. Appl. Cogn. Psychol. 24 (3),
312–325.
Kampmann, I.L., Emmelkamp, P.M., Morina, N., 2016. Meta-analysis of technology-as-
sisted interventions for social anxiety disorder. J. Anxiety Disord. 42, 71–84.
Linetzky, M., Pergamin‐Hight, L., Pine, D.S., Bar‐Haim, Y., 2015. Quantitative evaluation
of the clinical efficacy of attention bias modification treatment for anxiety disorders.
Depress. Anxiety 32 (6), 383–391.
Lopes, F.M., Viacava, K.R., Bizarro, L., 2015. Attentional bias modification based on vi-
sual probe task: methodological issues, results and clinical relevance. Trends
Psychiatry Psychother. 37 (4), 183–193.
MacLeod, C., Grafton, B., 2016. Anxiety-linked attentional bias and its modification: il-
lustrating the importance of distinguishing processes and procedures in experimental
psychopathology research. Behav. Res. Ther. 86, 68–86.
MacLeod, C., Mathews, A., 2012. Cognitive bias modification approaches to anxiety.
Annu. Rev. Clin. Psychol. 8, 189–217.
Mathews, A., MacLeod, C., 2005. Cognitive vulnerability to emotional disorders. Annu.
Rev. Clin. Psychol. 1, 167–195.
Menne-Lothmann, C., Viechtbauer, W., Höhn, P., Kasanova, Z., Haller, S.P., Drukker, M.,
Lau, J.Y., 2014. How to boost positive interpretations? A meta-analysis of the ef-
fectiveness of cognitive bias modification for interpretation. PLoS One 9 (6),
e100925.
Mogoaşe, C., David, D., Koster, E.H., 2014. Clinical efficacy of attentional bias mod-
ification procedures: an updated meta‐analysis. J. Clin. Psychol. 70 (12), 1133–1157.
Pergamin‐Hight, L., Pine, D.S., Fox, N.A., Bar‐Haim, Y., 2016. Attention bias modification
for youth with social anxiety disorder. J. Child Psychol. Psychiatry 57 (11),
1317–1325.
Price, R.B., Wallace, M., Kuckertz, J.M., Amir, N., Graur, S., Cummings, L., Bar-Haim, Y.,
2016. Pooled patient-level meta-analysis of children and adults completing a com-
puter-based anxiety intervention targeting attentional bias. Clin. Psychol. Rev. 50,
37–49.
See, J., MacLeod, C., Bridle, R., 2009. The reduction of anxiety vulnerability through the
modification of attentional bias: a real-world study using a home-based cognitive bias
modification procedure. J. Abnorm. Psychol. 118 (1), 65.
Sharpe, L., Ianiello, M., Dear, B.F., Perry, K.N., Refshauge, K., Nicholas, M.K., 2012. Is
there a potential role for attention bias modification in pain patients? Results of 2
randomised, controlled trials. Pain 153 (3), 722–731.
Sharpe, L., Johnson, A., Dear, B.F., 2015. Attention bias modification and its impact on
experimental pain outcomes: comparison of training with words versus faces in pain.
Eur. J. Pain 19 (9), 1248–1257.
Todd, J., Sharpe, L., Colagiuri, B., 2016. Attentional bias modification and pain: the role
of sensory and affective stimuli. Behav. Res. Ther. 83, 53–61.
White, L.K., Suway, J.G., Pine, D.S., Bar-Haim, Y., Fox, N.A., 2011. Cascading effects: the
influence of attention bias to threat on the interpretation of ambiguous information.
Behav. Res. Ther. 49 (4), 244–251.
Wiers, R.W., Gladwin, T.E., Hofmann, W., Salemink, E., Ridderinkhof, K.R., 2013.
Cognitive bias modification and cognitive control training in addiction and related
psychopathology: mechanisms, clinical perspectives, and ways forward. Clin.
Psychol. Sci. 1 (2), 192–212.