Systematic Review and Meta-Analysis: Eye-Tracking of

REVIEW

Systematic Review and Meta-Analysis: Eye-Tracking ofAttention to Threat in Child and Adolescent AnxietyStephen Lisk, MSc, Ayesha Vaswani, BSc, Marian Linetzky, MA,Yair Bar-Haim, PhD, Jennifer Y.F. Lau, PhD

Objective: Attention biases for threat may reflect an early risk marker for anxiety disorders. Yet questions remain regarding the direction and time-course of anxiety-linked biased attention patterns in youth. A meta-analysis of eye-tracking studies of biased attention for threat was used to compare thepresence of an initial vigilance toward threat and a subsequent avoidance in anxious and nonanxious youths.

Method: PubMed, PsycARTICLES, Medline, PsychINFO, and Embase were searched using anxiety, children and adolescent, and eye-tracking-related key terms. Study inclusion criteria were as follows: studies including participants �18 years of age; reported anxiety using standardized mea-sures; measured attention bias using eye tracking with a free-viewing task; comparison of attention toward threatening and neutral stimuli; and availabledata to allow effect size computation for at least one relevant measure. A random effects model estimated between- and within-group effects of firstfixations toward threat and overall dwell time on threat.

Results: Thirteen eligible studies involving 798 participants showed that neither youths with or without anxiety showed significant bias in first fixationto threat versus neutral stimuli. However anxious youths showed significantly less overall dwell time on threat versus neutral stimuli than nonanxiouscontrols (g ¼ L0.26).

Conclusion: Contrasting with adult eye-tracking data and child and adolescent data from reaction time indices of attention biases to threat, there wasno vigilance bias toward threat in anxious youths. Instead, anxious youths were more avoidant of threat across the time course of stimulus viewing.Developmental differences in brain circuits contributing to attention deployment to emotional stimuli and their relationship with anxiety are discussed.

Key words: child and adolescent anxiety, attention bias, threat processing, eye-tracking

J Am Acad Child Adolesc Psychiatry 2020;59(1):88–99.

A

88

nxiety disorders are a leading cause of morbidityglobally.1 As anxiety often starts in youth,2

identifying early risk markers that can be tar-
geted through interventions is a clear public health priority.Biased attention processing of threat may maintain anxiety3
and contribute to its onset,4 making it a potential treatmenttarget. Before the therapeutic potential of targeting atten-tion biases can be realized, outstanding questions on thepresence and direction of these biases in relation to youthanxiety need to be addressed, particularly as most studiesmeasuring attention biases and youth anxiety rely on indi-rect reaction time indices. Here, we present the first meta-analysis of newer studies using eye-tracking measures ofattention to assess bias for threat and their associations withyouth anxiety.

Effective detection of danger is fundamental to survival.However, cognitive accounts of anxiety suggest that someindividuals show exaggerated attentional processing ofthreat-related information, contributing to an interruption

www.jaacap.org

of daily functioning and manifesting as clinical anxiety.3

Indirect support for these models has emerged in adultsfrom tasks measuring reaction time (RT) under variousexperimental conditions. Most commonly used is the dot-probe task, a measure of biased visual spatial orienting to-ward (or away) from threat. Specifically, an attention biastoward threat (vigilance) is inferred by shorter RTs to detecta probe replacing a threatening stimulus than one replacinga nonthreatening stimulus, whereas an attention bias awayfrom threat (avoidance) is inferred by longer RTs to probedetection following a threatening versus nonthreateningstimulus. Using these tasks under short and long presenta-tion times of the threatening and nonthreatening stimuli,adult studies demonstrate anxiety-linked attention towardthreat at early involuntary stages of threat processing,consistent with facilitated threat-orienting (vigilance hypoth-esis).5-7 To a lesser extent, strategic avoidance of threat at laterstages of threat processing (vigilance-avoidance hypothesis)has also been reported.8 Another common task, which is

Journal of the American Academy of Child & Adolescent PsychiatryVolume 59 / Number 1 / January 2020

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jaac.2019.06.006&domain=pdf

http://www.jaacap.org

ATTENTION BIASES AND YOUTH ANXIETY

thought largely to tap the individual’s ability to disengageattention processing of distracting emotional informationrather than orienting, is the emotional Stroop task. Here,RTs are compared to color-naming threatening versusnonthreatening words. In adults, longer RTs to the formerversus latter condition have been found; together with datafrom other variations of the dot-probe task,9,10 these areinterpreted as being consistent with attentional maintenanceon threat in anxiety. Thus, adult RT data suggest thecontribution of varying components of attentional bias toanxiety.11

A significant number of studies have applied RT-basedtasks in children and adolescents with varying anxiety levels.Meta-analyzing 38 studies, Dudeney et al.12 found thatalthough both anxious and nonanxious youths displayedattention biases toward threat, anxious youths demonstrateda significantly greater bias than nonanxious youths. Thisbetween-group difference in bias increased with age acrossyouth; was greater when using the emotional Stroop taskthan the dot-probe task; and emerged only in dot-probestudies using a 1,250-millisecond presentation time ratherthan shorter presentation times (�500 milliseconds). Thus,although these data are clear in suggesting greater atten-tional maintenance on threat amongst anxious youths fromthe Stroop task, data from the dot-probe task are moremixed over whether there is also “early” vigilance for threatas reported in adults.5,11,13,14

One possibility why youth data vary from adult findingsis methodological. A longer time to process stimuli beforethe button press to maximize accuracy may be needed inyounger participants.12 Such concerns underscore the distalrelation between attentional processing and behavioralresponse. Indeed, RTs provide a relatively indirect mea-surement of attention.15 The resultant RT score does notaccount for variation in attentional processing after stimuluspresentation but before probe appearance, and factors suchas preparation and execution of motor response that mayvary between individuals, particularly children, canconfound RT differences.15 Furthermore, although RTtasks can potentially separate individual components un-derpinning attention bias (eg, facilitated threat orientingfrom attentional maintenance), they require multiple tasksand conditions to achieve this, which can cause fatigue andresponse errors in children.

Alternative approaches such as taking eye gaze measuresduring the presentation of threatening and nonthreateningstimuli16 have been used to more directly and continuouslymeasure attention.17 By measuring fixations (time andlocation of attentional deployment between saccadicmovements18), several components of attention bias can beassessed. Vigilance toward threat can be indexed by


recording the location of the individual’s first fixation afterstimulus presentation in each trial. Comparison of firstfixations to threat against first fixations to neutral provides aprobability score to indicate the direction of initial orient-ing. Greater probability of first fixation toward threat in-dicates a threat vigilance bias. Alternatively, the latency offirst fixations to each stimulus type can be compared: fasterfirst fixations to threatening versus neutral stimuli couldindicate a vigilance bias. To measure maintained attentionon threat across stimulus presentation time, dwell time oneach stimulus type is calculated across the entire trial.Greater mean dwell time on threatening stimuli than onnonthreatening stimuli indicates maintained attention to-ward threat, with the opposite suggesting overall threatavoidance. Attentional vigilance and avoidance patterns overtime can also be derived through dwell time on threateningand nonthreatening stimuli measured across time windows(epochs).

Meta-analyses of eye-tracking data from adults15 showthat during free viewing and visual search tasks, anxiousadults demonstrated greater initial vigilance for threatcompared to nonanxious adults, consistent with RTstudies.5 Total dwell time on threat versus nonthreat stimuliover the entire stimulus presentation time was not investi-gated in this meta-analysis. Developmental differences inbrain circuits contributing to attention deployment in thepresence of emotional stimuli19 and, more particularly, inhow anxiety-linked attention biases change with matura-tion12,20 means that adult findings cannot be extrapolatedto youth populations. Therefore, a growing number of eye-tracking studies of threat processing in youths have beenconducted. Mixed findings across studies warrants poolingthese data to evaluate combined effect sizes of anxiety-linkedbiases. Most studies have used free-viewing tasks, as thesereflect a more ecologically valid assessment of attentiondeployment but also, more importantly, may be suitable forchildren and adolescents, as they are less dependent on taskperformance, which could introduce age-associated con-founds. Most studies have investigated vigilance to threat bymeasuring probability of first fixation toward threat, andmaintained attention on threat, through total dwell time onthreatening/nonthreatening stimuli.

This meta-analysis aimed to address the followingquestions: First, do anxious children/adolescents and theirnonanxious counterparts show an absolute bias (signifi-cantly different from zero) in probability of first fixation tothreatening stimuli (as an index of initial threat-vigilance),and is there a between-group difference on this measure?Second, do anxious children/adolescents and their non-anxious counterparts show an absolute bias (significantlydifferent from zero) in total dwell time on threatening

www.jaacap.org 89


LISK et al.

versus neutral stimuli (as an index of maintained attention),and is there a between-group difference on this measure? Asresearchers have addressed these questions using differenttask parameters and recruiting specific subpopulations,which could affect findings, we investigated the effects ofvarious procedural and population moderators. (1) Primaryattention task: as these have varied between a free-viewingapproach and dot-probe tasks that contain additionalactive components of probe selection resulting in antici-patory eye movements during the free-viewing element, thismay affect first fixation results.21 (2) Stimulus presentationtime: as brief presentation times are thought to captureinvoluntary attentional deployment and longer viewingtimes, more strategic processes,13 analyses of total dwelltime from studies using different viewing times may beprone to differential influences of involuntary and strategicprocesses, affecting the presence and direction of the bias.(3) Age: as attention biases vary with age,12 coinciding withdevelopmental accounts that suggest that all children beginwith an attention bias toward threat, which “corrects”during healthy developmental trajectories,22 findings fromstudies using different age ranges (eg, child/adolescent) maycapture distinct attentional response to threat. (4) Clinicaldiagnosis: as some studies have found a threat-bias amongclinically anxious youths only,23,24 symptom severity couldmodify the expression of the attention bias. (5) Anxietysubtype: as many studies investigating attention biases usesamples containing individuals with mixed anxiety di-agnoses or features, this may alter the intensity of the threatstimuli used in the tasks across participants.

METHODEligibility CriteriaWe included studies that met the following criteria: (1)Because of practical considerations relating to translation,and because the majority of biomedical literature is pub-lished in English-language journals,25 with no clear sys-tematic bias of such language restriction in trials reported inconventional medicine,26 the study had to be available inEnglish. (2) The study must be an original investigation. (3)The study must investigate human participants �18 yearsof age. (4) The study must use a standardized measure ofanxiety (state or trait) for all participants, either clinicalinterview or a self/parent-report anxiety questionnaire. (5)The study must use eye tracking to measure attention bia-ses. (6) The study must use a free-viewing task, or a taskwith a free-viewing element (such as dot-probe), duringwhich gaze is tracked. As these tasks are less dependent ontask performance, this minimizes age-associated confoundsin studies of children and adolescents. (7) Appropriate datamust be available to compute an effect size for at least one of

90 www.jaacap.org

the bias measures being investigated (probability of firstfixation to threat; latency of first fixation to threat; totaldwell time on threat versus neutral). These data may beavailable as mean scores for “anxious” and “nonanxious”groups, a test statistic for group difference, or a correlationbetween the attention measure and anxiety severity. If dataare unavailable in the paper, they must be made available bythe author. (8) The design must allow for the comparison ofattention toward threatening and neutral elements of thearray. Studies pairing threatening stimuli with stimuli of anyother valence were excluded (such as one that paired fearand angry faces with a mixture of happy and neutralfaces27).

Information Sources and Search TermsIn April 2018, PubMed, Psycharticles, Medline, Psychinfo,and Embase databases were searched for eligible studies. Weused anxiety-related key terms (anx*, anxiety disorder,GAD, depress*, fear, phobi*, dysphori*, and panic) thatwere crossed with key terms for eye-tracking measures (eye*,gaze*, fixation*, dwell time, and saccade) and key terms toidentify child and adolescent participants (child*, adol*,pediatric, youth, juvenile, and teen*). Reference lists ofidentified studies were examined further for potentiallyeligible research, as were any identified relevant review pa-pers. Titles and abstracts were screened for inclusion by theauthors (S.L., A.V., J.L.) based on eligibility criteria 1through 5. There was 100% agreement across authors.Studies that met this eligibility criterion were retained forfull-text review to assess whether they met all criteria (S.L.,A.V.), again with 100% agreement. Where studies met allinclusion criteria but further data were required, authorswere contacted to request the necessary data.

Statistical AnalysisResearch Questions and OutcomeMeasures. Meta-analyseswere carried out to test two questions. First, the vigilancehypothesis was examined, namely, that individuals with ananxiety disorder would detect threat more readily, and thusorient to it more often, than nonanxious controls. Thevigilance hypothesis was investigated using studies thatrecorded the direction of initial gaze orienting; specifically,measures of probability of first fixation to threat versusneutral stimuli and latency of first fixation toward threat-ening stimuli were used. Studies that did not report firstfixation probability or latency but reported only total fixa-tion time on threatening stimuli in the first 500þ milli-seconds were excluded from the analysis (k ¼ 2). Second,we tested the maintenance hypothesis, namely, that anxietyis characterized by maintained attention on threat; thus,across the entire trial, individuals with anxiety will more




often dwell upon threatening than neutral stimuli. Thishypothesis was investigated using studies that recordedthe mean duration of gaze (dwell time) on threat versusneutral stimuli, when stimuli were displayed for >1,000milliseconds.

Data Coding. Studies were coded on the following vari-ables: number of participants; participants’ mean age; sexsplit (% female participants); sample type (clinical/analogue); type of anxiety disorder; experimental task (free-viewing/dot-probe/other); type of threat stimulus (face/picture); threat emotion; number of stimuli presented; andstimulus presentation time. When the study included resultsfrom “with/without stressor” groups separately, data fromthe “without stressor” condition were used to retain con-sistency across the sample (k ¼ 2).

Analyses. With our measures of first fixation probability orlatency, and overall dwell time, we performed within-groupanalyses for the anxious and nonanxious groups to evaluatefor the presence of an “absolute” bias toward threat in eithergroup. We also estimated the between-group differencebetween anxious and nonanxious individuals on first fixa-tion measures. Because of a lack of relevant data for within-group analysis of dwell time, only between-group analysiscould be carried out for this attention bias measure.

Meta-analyses were conducted using ComprehensiveMeta-Analysis software (version 3.3.070). All effect sizeswere calculated using Hedges’ g. To interpret effects withthis measure, Cohen’s d guidelines were used28 (small effect:0.20; moderate effect: 0.50; large effect: 0.80). For thebetween-group analysis of both first fixation data and overalldwell time, effect size direction was calculated so that apositive effect size indicates that the attentional bias towardthreat is larger in anxious participants than in control par-ticipants. In studies that did not use high- and low-symptom groups, correlations between symptom severityand attention bias were used, with a positive effect sizeindicating a greater attention bias toward threat for moreanxious individuals. In the within-group analyses, a positiveeffect size indicates that the attentional bias is greater forthreat stimuli than for neutral stimuli, with a negative effectsize indicating the opposite. A random-effects model waschosen to compute combined effect sizes, as heterogeneitywas expected across studies, and this method allows theresults to be generalized to similar studies.29 To assessheterogeneity of overall effect sizes, Cochran’s Q30 was used.In addition, the I2 statistic31 was used, indicating the per-centage of this variation across effect sizes.

Categorical variables were identified as potential mod-erators, consisting of procedural and population factors thatdiffered across studies. Procedural variables included the


attention task (dot-probe or free-viewing) and the stimuluspresentation time (�2,000 milliseconds or >2,000 milli-seconds). Population variables included the following: agegroup (adolescent, mean age of �12 years; or child, meanage <12 years); sample type (clinical or analogue); anxietytype (social anxiety disorder/social phobia or mixed, morethan one anxiety disorder included). Moderator analyseswere conducted in relation to outcomes on between-groupmeasures of first fixation and dwell time. Because of thesmall number of studies eligible for within-group analysis,moderator analysis was not carried out for the within-groupresults.

Risk of Publication Bias. Funnel plots were inspected for allanalyses to assess publication bias. Rank correlation32 andregression tests33 were also carried out to evaluate evidenceof publication bias, as well as the Duval and Tweedie trimand fill method.34 Fail-safe numbers were computed toassess the magnitude of a potential file-draw problem. Thisprovides an estimate of the number of studies, with an effectsize of zero that would need to be added to the analysis toproduce a cumulative effect that is statistically nonsignifi-cant (p > .05). In addition, we used the Orwin fail-safe N35

to calculate the number of studies with an effect size of zerothat would need to be added to the analysis to produce aspecified “trivial” Hedges’ g value.

RESULTSSearch ResultsFigure 1 illustrates the literature search and study selectionprocess. Initial searches identified 3,871 studies. Afterremoving duplicates, this was reduced to 1,818 studies.After excluding by abstract, this number was reduced to 29studies. Full-text screening resulted in exclusion of 16 morestudies, resulting in 13 eligible studies.

Study CharacteristicsStudy characteristics are displayed in Table 1.36-48 Theentire data set was scanned for outliers; these were iden-tified as studies in which 95% CIs did not overlap with the95% CI of the combined effect size. No studies yielded aneffect size that was an outlier. Therefore, the total sampleincluded data from 798 participants aged 3 to 18 years,from 13 studies. All studies were published in peer-reviewed journals. Although they contained a free-viewing element, studies varied on the specific tasksused; nine studies used a task that solely involved freeviewing of the presented stimuli, whereas four studies useda dot-probe task that required a user action after the periodof free viewing. Nine studies used a clinical sample ofanxious participants, and four studies used an unselected

www.jaacap.org 91


LISK et al.

sample. Five studies investigated attention bias in relationto social anxiety disorder (SAD) or social phobia (SP), twoused broader overall anxiety scores, one for state anxiety,and the remaining five included patients with a mixture ofanxiety diagnoses (including SAD, SP, generalized anxietydisorder [GAD], and separation anxiety [SEP]). Tenstudies used faces as the threatening stimuli, with five ofthese using an angry emotion, three using fear, one usingpain, and one specifying a general “threatening” face.Two studies used eyes as the threatening stimuli, one aspart of the face, and the other study only the eyes. Thefinal study used pictures of social scenes, with faceswithin the scenes defined as the threatening stimuli.Effect sizes within each study, and CIs can be seen inFigures 2 to 4.

Meta-Analysis of Anxiety and First Fixation DataWithin-Group Analyses. The meta-analyses examiningwithin-group differences in first fixation on threat versusneutral stimuli (Figure 2) show that the combined effect sizewas not significant in anxious participants (k ¼ 6; g ¼0.315, p ¼ .21, CI ¼ �0.17, 0.80), or in nonanxiouscontrols (k ¼ 6; g ¼ 0.27, p ¼ .27, CI ¼ �0.21, 0.75).

FIGURE 1 Flowchart of Screening Processes for StudyInclusion

Note: Criterion 4 did not use standardised measure of anxiety; criterion 6 did notuse appropriate attention task; for Criterion 7, necessary data were unavailable/un-obtainable; criterion 8 did not allow for comparison of attention toward threat-ening and neutral stimuli.

92 www.jaacap.org

There was large heterogeneity in the effect sizes for anxious[Q(5) ¼ 46.32, p < .001, I2 ¼ 89.20%] and nonanxious[Q(5) ¼ 39.48, p < .001, I2 ¼ 87.33%] groups.

Between-Group Analysis. The meta-analysis examiningthe between-group differences in first fixation on threat(Figure 3) found that anxious individuals did not signifi-cantly differ from nonanxious individuals in initial fixationtoward threatening versus neutral stimuli (k ¼ 8; g ¼ 0.04,p ¼ .39, CI ¼ �0.18, 0.26). There was not significantheterogeneity in the effect sizes [Q(8) ¼ 8.56, p ¼ .29,I2 ¼ 18.25%].

Meta-Analysis of Anxiety and Overall Dwell TimeBetween-Group Analysis. The overall effect size for themeta-analysis examining the association between anxietyand dwell time (Figure 4) was significant (k ¼ 12;g ¼ �0.26, p ¼ .004, CI ¼ �0.44, �0.08), indicating thatanxious youths avoided threatening stimuli more thannonanxious youths across the stimulus-viewing period.There was not significant heterogeneity in the effect sizes[Q(11) ¼ 15.48, p ¼ .16, I2 ¼ 28.93%]. Of note, becauseresults from analogue samples can be difficult to interpret,as nonclinical individuals with high scores on self-reportedanxiety measures do not always show patterns of attentionsimilar to those of clinical patients, we re-ran the analysisexcluding analogue studies. Re-running the between-groupanalysis on dwell time data without analogue samples stillshowed the overall effect [k ¼ 8; g ¼ �0.24, p ¼ .035,CI¼ �0.45,�0.02); heterogeneity Q(11)¼ 9.97, p¼ .19,I2 ¼ 29.77%].

Subgroup Moderator analysesThe nonsignificant c2 values in testing for heterogeneity invariance, and I2 values that are not extremely high, suggestthat the studies in each sample were fairly homogeneous.However, as the I2 values were approaching 25%, and,based upon a priori analysis plans, moderator analyses wereconducted.

Subgroup Moderator Results for Between-Group Com-parisons of First-Fixation Data. There were no significantmoderation effects on first-fixation data by population orprocedural factors identified a priori (Table 2).

Subgroup Moderator Results for Between-Group Com-parisons of Overall Dwell Time. For anxiety type, signifi-cantly greater (negative) between-group effect sizes(indicating more avoidance of threat for anxious comparedto nonanxious individuals) was found for studies includingparticipants with a mixture of anxiety types than for studiesusing only social anxiety (p ¼ .05) (Table 2).



TABLE 1 Study Characteristics

Study Nn

(Clinical)n

(Control)Age,

Range (y)Age,

mean (y)

%Sex

(Female)SampleType

PrimaryAnxietyProblem

AttentionTask

ThreatStimulus

ThreatEmotion

Number ofStimuli

DisplayTime (ms)

Capriola-Hall et al.(2018)36

41 41 N/A Adolescents(12L16)

14.54 68% Clinical SAD Free-viewing

Face Angry 2 3,000

Dodd et al. (2015)37 83 37 46 Children(3L4)

3.99 59% Clinical SAD, GAD,SEP, SP

Free-viewing

Face Angry 2 1,250

Haller et al. (2017)38 51 N/A N/A Adolescents(14L18)

16.73 100% Analogue SAD Free-viewing

Scene Social Varying 5,000

Heathcote et al.(2016)39

37 N/A N/A Adolescents(8L17)

12.1 64% Analogue State anxiety Free-viewing

Face Pain 2 3,500

Kleberg et al.(2017)40

25 25 N/A —

Adolescents15.2 84% Clinical SAD Free-

viewingEyes Eyes 4 2,000

Michalska et al.(2017)41

82 N/A N/A Children(9L13)

11.81 60 % Analogue Overallanxietyscore

Free-viewing

Face Eyes 1 7,000L8,000

Price et al. (2013)42 94 74 20 Children—

10.57 52% Clinical GAD, SEP, SP Dot-probe

Face Fear 2 2,000

Price et al. (2016)43 67 67 N/A Children(9L14)

11.1 53.7% Clinical GAD, SEP, SP Dot-probe

Face Fear 2 2,000

Schmidtendorfet al. (2018)44

79 37 42 Children—

11.45 61% Clinical SAD Free-viewing

Face Angry 2 5,000

Seefeldt et al.(2014)45

73 30 43 Children(8L12)

9.9 44% Clinical SP Dot-probe

Face Angry 2 3,000

Shechner et al.(2013)46

33 18 15 Adolescents(8L17)

13.19 58% Clinical GAD, SAD, SP Free-viewing

Face Angry 2 10,000

Shechner et al.(2017)47

45 19 26 Adolescents(8L17)

12.63 44% Clinical GAD, SAD, SP Free-viewing

Face Threat 3 5,000

Tsypes et al.(2017)48

88 N/A N/A Children—

9.26 44% Analogue Overallanxietyscore

Dot-probe

Face Fear 2 1,000

Note: GAD ¼ generalized anxiety disorder; SAD ¼ social anxiety disorder; SEP ¼ separation anxiety disorder; SP ¼ social phobia (spider).

Journalofthe

American

Acad

emyof

Child

&Adolescent

Psychiatrywww.jaacap

.org93

Volum

e59

/Num

ber

1/January

2020

ATTEN

TION

BIASES

AND

YOUTH

ANXIETY


FIGURE 2 Forest Plot of Within-Group First Fixation Bias for Threatening Stimuli

Note: 95% CIs and study weights illustrate contribution to overall effect size. Diamond represents estimate of combined effect size.

LISK et al.

Publication BiasFunnel plots were inspected (see Figures S1 and S2, availableonline), and no evidence of asymmetry was observed. TheEgger test33 and rank correlation test32 results were allnonsignificant (all p values >.49). Furthermore, using theDuval and Tweedie trim and fill procedure,34 no evidence ofpublication bias was found for any of the measures. For thedwell time meta-analysis, the fail-safe N49 was 25, meaningthat there would need to be 25 studies with an effect size of

FIGURE 3 Forest Plot of Between-Group First Fixation Bias for T

Note: 95% CIs and study weights illustrate contribution to overall effect size. Diamond

94 www.jaacap.org

0 added to the analysis to increase the p value to>.05, that is,to produce a statistically nonsignificant cumulative effect. Inaddition to this, using the Orwin fail-safe N, to bring ourcriterion down to a Hedges’ g value of�0.1, it would take 21extra studies with an effect size of 0.

DISCUSSIONThis first meta-analysis of eye-tracking measures of atten-tion bias in child and adolescent anxiety included data from

hreatening Stimuli

represents estimate of combined effect size.



FIGURE 4 Forest Plot of Between-Group Dwell Time Bias for Threatening Stimuli

Note: 95% CIs and study weights illustrate contribution to overall effect size. Diamond represents estimate of combined effect size.


798 participants aged 3 to 18 years across 13 studies. Asignificantly greater tendency to direct first fixations onthreatening over neutral stimuli did not characterize ordifferentiate anxious and nonanxious children or adoles-cents. Instead, anxious youths showed a greater tendency toavoid maintaining their gaze on threat compared to non-anxious youths, a difference that emerged only in studies inwhich samples comprised mixed anxiety diagnoses.

At first glance, our findings that biased orienting towardthreat did not differentiate anxious and nonanxious childrenand adolescents but that over the course of stimulusviewing, anxious youths avoided threatening overnonthreatening stimuli more than nonanxious controlsseems incompatible with the meta-analyses of RT data inchildren and adolescents.12 However, it is possible that firstfixation data may not be equivalent to attention capture/engagement in RT-based paradigms. There is therefore stillwork to do in mapping how RT-based indices relate to eye-tracking indices. Closer inspection of the moderator analysisin the meta-analyses also shows that anxiety group differ-ences for attention bias emerged only only when stimuli indot-probe tasks were presented at 1,250 milliseconds (ratherthan those >500 milliseconds) and only when they weregreater than dot-probe when considering emotional Strooptask results. These data could be interpreted to imply thatany bias in attentional deployment for threat in youths islikely to occur beyond initial fixation and is driven by dis-turbances in voluntary top-down processes. Our findingsextend these interpretations by suggesting that these later-stage biases result in an eventual strategic avoidance of


threat. As our total dwell time score is unable to inferspecific patterns of attention bias over time, it is less clearwhether the direction of these biases indeed fluctuate acrosstime in anxious versus nonanxious youths. In-Albon et al.found a pattern of vigilance-avoidance in anxious youth intwo studies,50,51 although a third study52 failed to reportsimilar patterns. These conflicting data underscore a needfor more studies using time windows with fixations anddwell time across different epochs. There are alternativemeasures derived from eye tracking that can be used toassess anxiety-linked differences across the entire viewingperiod of complex stimuli,53 such as assessing the visual scanpath, a sequence of fixations and saccades thought to reflectthe manner in which information is attended to, reap-praised, and integrated.

Second, our meta-analytic findings appear inconsistentwith adult RT5 and eye-tracking data15 which suggest thatanxiety is characterized by facilitated detection and orient-ing of initial attention toward threat and greater maintainedattention on threat in anxious individuals.54-58 These dif-ferences may instead underscore the importance of devel-opmental accounts of anxiety. Such accounts need torecognize greater variability in attention bias expressionamong young people compared to adults, manifesting be-tween initial vigilance, rapid avoidance, sustained threatmonitoring, and vigilance-avoidance patterns, and may beattributable to the influences of multiple cognitive andlearning processes unique to the developing individual.59

However, that developmental factors may moderate atten-tion bias expression across youth was not supported by our

www.jaacap.org 95


TABLE 2 Moderator Results for Between-Group Comparisons of Attentional Vigilance and Attentional Maintenance

Effect size Heterogeneity Moderation

Moderators for Between-Group Comparisons ofAttentional Vigilance k g 95% CI I2 Q pAge groupAdolescent 3.00 0.15 e0.62, 0.93 70.27 0.20 .66Child 5.00 e0.03 e0.25, 0.2 0.00Presentation Time, ms<2,001 4.00 e0.07 e0.33, 0.19 0.00 1.36 .24>2,000 4.00 0.20 e0.18, 0.58 32.90TaskDot-probe 4.00 0.01 e0.26, 0.28 0.00 0.06 .81Free-viewing 4.00 0.08 e0.44, 0.6 63.17Anxiety TypeMixed 4.00 0.21 e0.11, 0.54 26.71 2.61 .11SAD/SP 4.00 e0.14 e0.43, 0.15 0.00

Moderators for Between-Group Comparisons ofAttentional Maintenance

Effect size Heterogeneity Moderation

k g 95% CI I2 Q PAge groupAdolescent 5 e0.19 e0.61, 0.22 49.82 0.20 .653Child 7 e0.30* e0.48, e0.11 14.22Presentation Time, ms<2,001 5 e0.35* e0.60, e0.16 0 1.37 .242>2,000 7 e0.16 e0.45, 0.13 50.47TaskDot-probe 4 e0.24 e0.57, 0.09 49.04 0.02 .881Free-viewing 8 e0.27* e0.5, e0.05 26.70Sample TypeAnalogue 4 e0.30 e0.63, 0.04 41.35 0.07 .791Clinical 8 e0.24* e0.46, e0.02 29.77Anxiety TypeMixed 6 e0.43* e0.63, e0.24 0 3.83* .050SAD/SP 5 e0.08 e0.37, 0.21 14.07

Note: The number of studies using an analogue group (k ¼ 0) was not enough to test moderation of “sample type.” Mixed ¼ studies includingpatients with a range of anxiety diagnoses; SAD ¼ Social Anxiety Disorder; SP ¼ Social Phobia.*p < .05.

LISK et al.

data. We found no within-group vigilance effect in anxiousor nonanxious children and adolescents, and no moderatingeffect of age on between-group differences in vigilance withinthis age range. These data therefore seem to speak againstdevelopmental accounts that all children may begin with anattention bias toward threat, which then “corrects” duringhealthy developmental trajectories.22 However caution isneeded before drawing firm conclusions. Albeit not reachingsignificance as a moderator, when categorizing the studies byage, there were differences in the strength of the associationbetween dwell time on threat and anxiety across age groups:the child group studies showed a significant avoidance,

96 www.jaacap.org

whereas the adolescent group studies did not. This is sur-prising, as the literature proposes avoidance as a maladaptiveemotion regulation strategy, driven largely by executivecontrol processes that mature in youth.60,61 Instead, puttingour findings with those from adults suggests that althoughattentional avoidance of threat characterizes anxious childrenand difficulty disengaging from threat characterizes anxiousadults, there are no clear attentional strategies in anxiousadolescents, possibly as brain circuits underlying attentionaldeployment are still undergoing reorganization duringadolescence. However, there was a relatively high heteroge-neity of variance between effect sizes in the adolescent group,




indicating that other factors may be affecting the associationsbetween anxiety and avoidance. Furthermore, as manystudies used wide age ranges, we could not investigate theinfluence of age through a meta-regression. Instead, we reliedon subgroup analysis, which crudely used mean age of thesample to dichotomously categorize studies into children andadolescents. Further research assessing the association be-tween anxiety-linked attention patterns across specific ages(or developmental stages) within youth would help toelucidate the role of maturation and/or experience on theexpression of these biases.

Finally, the only factor that significantly moderatedmaintained attention was “anxiety type”; only studies usingparticipants with a mixture of anxiety types found a sig-nificant between-group difference in avoidance. It should benoted, however that studies using mixed anxiety groups allincluded patients with social anxiety within their samples;plus, given high level of homotypic comorbidity in anxietydisorders,62 several of the “only” social anxiety studies mayhave included co-occurring anxiety disorders, making itdifficult to disentangle biases in maintained attention perdisorder. However, as a whole, the results imply that specificdiagnostic subgroups other than social anxiety disorder aredriving this avoidance effect. Using more specific disorderand symptom boundaries in future study designs may bemore informative as attentional components increasinglyshow disorder and symptom specificity.63,64

There are several limitations to our study. Compared toother meta-analyses of attention bias to threat5,12,15, therewere fewer studies in this meta-analysis, with a handful ofpublished studies that were excluded because of inadequateand unavailable data to compute an effect size. Null resultsmay have emerged from low power. A low number ofstudies also prevented some moderator analyses from beingcarried out, and others being considered (such as a meta-regression of the ratio of female-to-male participants inthe sample), as well as the presence of comorbidity withnonanxiety disorders in some samples (eg, autism spectrumdisorders or depression). For one moderator, presentationtime, the rationale for selecting a cut-off (2,000 millisec-onds) was somewhat arbitrary, driven by the distribution ofthe parameters used in individual studies, and the need toachieve a largely even split of studies into long and shortdurations (five versus seven). Where moderation wasexamined, differential effects across levels of some variablesmay have reached significance with larger samples.

Second, although first-fixation data via eye trackingprovide a more precise indication of where overt attention isfirst directed, it is unclear whether these measures in factreflect a mixture of stimulus-driven and strategic processes.Previous research has found typical latency of exogenous


first fixations to be around 175 milliseconds,52 whereas eye-tracking studies from anxious individuals generally showfirst-fixation latency to be longer (250�400 ms16,65). Thereare also suggestions first fixation measures of threat pro-cessing aren’t as reliable as expected,54,66,67 and may beaffected by participants favoring fixation to the top or leftimage regardless of its emotional valence. Relatedly, thefree-viewing approach across long stimulus duration times(>1,000 milliseconds) used by many eye-tracking studiesmay have an impact on identifying anxiety group differencesin first fixations. As this task measures only spontaneousviewing behavior, and not attentional behavior related totask demands, it may be less powerful in tapping attentionalengagement/disengagement, as neither is necessary for taskcompletion. Indeed, group differences in attention bias aremore readily identified when a task action is required, suchas a visual search task.68-70 Another way of potentiallyinforming the attentional components contributing toanxiety is to simultaneously collect pupillary dilation data.Future studies could try to associate these measures and togain information on the online interplay between thetemporal dynamics of gaze behavior and brain-mediatedemotional responsiveness.

Finally, many studies may not always accurately identifybiases in relation to threat due to differences in threatevaluations. For instance, all facial stimuli may be consid-ered somewhat threatening in socially anxious individuals,and as such avoidance of all faces may occur.71 Avoidance ofall perceived social threat may mask any group differences inattention bias picked up with current measures, as onlybetween-face differences are generally calculated. It couldalso be possible that no threat evaluation occurs becausesuch threat stimuli lack personal relevance to young people,and that this also masks within or between-group anxietydifferences in attention bias.

Notwithstanding these limitations there are someclinical implications of these findings. Based on relativelyrobust findings from anxious adults of an association withattention bias for threat,5 attention bias modification(ABM) tasks have been used as anxiety-reducing in-terventions in adults72,73 and young people,74 mainlyusing the dot-probe task but also eye-tracking tasks.75,76

These paradigms train attention away from threatmostly toward nonthreatening stimuli. Results usingABM in anxious youth have been mixed, with meta-analyses finding that ABM did not lead to a signifi-cantly greater symptom reduction than a control condi-tion.77 The current meta-analysis results suggest thatrather than modify an initial orienting bias for threat itmay be valuable to modify strategic processes that reducethreat avoidance. Some studies have already suggested

www.jaacap.org 97


LISK et al.

that ABM reduces anxiety by improving strategic atten-tion control processes,78 and, within this, some theoristssuggest that visual search tasks may be more appropriatefor modifying these voluntary aspects of attention.13

Indeed, in youth, implementations of visual searchtasks, where participants search for a benign target(smiling face) from among negative distractors (negativefaces) has resulted in consistent symptom reduction,79,80

although it remains to be seen whether this reduction canbe explained by threat avoidance specifically, rather thanexposure to threatening faces per se.

In summary, the current meta-analyses suggest thatanxious and non-anxious youth do not differ in overt initialorienting to threat, as measured by eye movements; how-ever, our results demonstrate a small effect of anxious youthavoiding threat. Future research with large sample sizes isrequired to investigate the avoidant pattern of strategicattention across time more discretely, and to delineate thefactors contributing to the individual differences found inattention bias expression amongst anxious youth.

98 www.jaacap.org

Accepted June 26, 2019.

Mr. Lisk, Ms. Vaswani, and Dr. Lau are with Institute of Psychiatry, Psychologyand Neuroscience, King’s College London. Ms. Linetzky and Dr. Bar-Haim arewith the School of Psychological Sciences, Tel Aviv University, Israel. Dr. Bar-Haim is also with the Sagol School of Neuroscience, Tel Aviv University, Israel.

Mr. Lisk is supported by a UK Medical Research Council studentship (MR/K50130X/1) and the European Commission FP7 Braintrain grant (602186).Dr. Lau has received funding from the UK Medical Research Council (MR/N006194/1).

This article is part of a special series devoted to the subject of anxiety andOCD. The series covers current topics in anxiety and OCD, including epide-miology, translational neuroscience, and clinical care. The series was edited byGuest Editor Daniel A. Geller, MBBS, FRACP.

Disclosure: Dr. Bar-Haim has active competitive grants from the U.S. Depart-ment of Defense, the US-Israel Binational Science Foundation, the Israel Sci-ence Foundation, and JOY Ventures. Dr. Lau has active competitive grantsfrom the British Academy and Mental Health Research UK. Mr. Lisk, Ms. Vas-wani, and Ms. Linetzky have reported no biomedical financial interests or po-tential conflicts of interest.

Correspondence to Jennifer Y.F. Lau, PhD, Department of Psychology, Insti-tute of Psychiatry, Psychology and Neuroscience, King’s College London, DeCrespigny Park, London SE5 8AF, UK; e-mail: [email protected]

0890-8567/$36.00/ª2019 American Academy of Child and AdolescentPsychiatry

https://doi.org/10.1016/j.jaac.2019.06.006

REFERENCES
1. World Health Organization. Depression and Other Common Mental Disorders: Global
Health Estimates. Geneva: World Health Organization; 2017.2. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. Lifetime

prevalence and age-of-onset distributions of DSM-IV disorders in the National Co-morbidity Survey Replication. Arch Gen Psychiatry. 2005;62:593-602.

3. MacLeod C, Mathews A, Tata P. Attentional bias in emotional disorders. J AbnormPsychol. 1986;95:15-20.

4. Lau JY, Waters AM. Annual research review: An expanded account of information-processing mechanisms in risk for child and adolescent anxiety and depression. J ChildPsychol Psychiatry. 2017;58:387-407.

5. Bar-Haim Y, Lamy D, Pergamin L, Bakermans-Kranenburg MJ, van IJzendoorn MH.Threat-related attentional bias in anxious and nonanxious individuals: a meta-analyticstudy. Psychol Bull. 2007;133:1-24.

6. Beck AT, Clark DA. An information processing model of anxiety: automatic and strategicprocesses. Behav Res Ther. 1997;35:49-58.

7. Eysenck MW, Derakshan N, Santos R, Calvo MG. Anxiety and cognitive performance:attentional control theory. Emotion. 2007;7:336-353.

8. Mogg K, Bradley BP, Miles F, Dixon R. Brief report time course of attentional biasfor threat scenes: testing the vigilance-avoidance hypothesis. Cogn Emot. 2004;18:689-700.

9. Fox E, Russo R, Bowles R, Dutton K. Do threatening stimuli draw or hold visualattention in subclinical anxiety? J Exp Psychol Gen. 2001;130:681-700.

10. Fox E, Russo R, Dutton K. Attentional bias for threat: evidence for delayed disengage-ment from emotional faces. Cogn Emot. 2002;16:355-379.

11. Cisler JM, Koster EH. Mechanisms of attentional biases towards threat in anxiety dis-orders: an integrative review. Clin Psychol Rev. 2010;30:203-216.

12. Dudeney J, Sharpe L, Hunt C. Attentional bias towards threatening stimuli in childrenwith anxiety: a meta-analysis. Clin Psychol Rev. 2015;40:66-75.

13. Mogg K, Bradley BP. Anxiety and attention to threat: cognitive mechanisms andtreatment with attention bias modification. Behav Res Ther. 2016;87:76-108.

14. Williams JMG, Watts FN, MacLeod C, Matthews A. Cognitive Psychology andEmotional Disorders. Oxford, UK: John Wiley & Sons; 1988.

15. Armstrong T, Olatunji BO. Eye tracking of attention in the affective disorders: a meta-analytic review and synthesis. Clin Psychol Rev. 2012;32:704-723.

16. Garner M, Mogg K, Bradley BP. Orienting and maintenance of gaze to facial expressionsin social anxiety. J Abnorm Psychol. 2006;115:760-770.

17. Duchowski AT. Eye tracking methodology: theory and practice. London: Springer-Verlag London; 2007.

18. Salvucci DDG, J. H. Identifying fixations and saccades in eye-tracking protocols. Pro-ceedings of the 2000 Symposium on Eye Tracking Research & Applications. ACM.2000;71-78.

19. Scerif G. Attention trajectories, mechanisms and outcomes: at the interface betweendeveloping cognition and environment. Dev Sci. 2010;13:805-812.

20. Cohen Kadosh K, Heathcote LC, Lau JY. Age-related changes in attentional controlacross adolescence: how does this impact emotion regulation capacities? Front Psychol.2014;5:111.

21. Rommelse NN, Van der Stigchel S, Sergeant JA. A review on eye movement studies inchildhood and adolescent psychiatry. Brain Cogn. 2008;68:391-414.

22. Field AP, Lester KJ. Is there room for ’development’ in developmental models of in-formation processing biases to threat in children and adolescents? Clin Child Fam Psy-chol Rev. 2010;13:315-332.

23. Waters AM, Henry J, Mogg K, Bradley BP, Pine DS. Attentional bias towards angryfaces in childhood anxiety disorders. J Behav Ther Exp Psychiatry. 2010;41:158-164.

24. Waters AM, Mogg K, Bradley BP, Pine DS. Attention bias for angry faces in childrenwith social phobia. J Exp Psychopathol. 2011;2:jep-018111.

25. Bown MJ, Sutton AJ. Quality control in systematic reviews and meta-analyses. Eur JVasc Endovasc Surg. 2010;40:669-677.

26. Morrison A, Polisena J, Husereau D, et al. The effect of English-language restriction onsystematic review-based meta-analyses: a systematic review of empirical studies. Int JTechnol Assess Health Care. 2012;28:138-144.

27. Lagattuta KH, Kramer HJ. Try to look on the bright side: children and adults can(sometimes) override their tendency to prioritize negative faces. J Exp Psychol Gen.2017;146:89-101.

28. Cohen J. Statistical Power Analysis for the Behavioral Sciences. Second ed. Hillsdale, NJ:Erlbaum; 1988.

29. Field AP. Meta-analysis of correlation coefficients: a Monte Carlo comparison of fixed-and random-effects methods. Psychol Methods. 2001;6:161-180.

30. Hedges L, Olkin I. Statistical Methods for Meta-analysis. San Diego, CA: AcademicPress; 1985.

31. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med.2002;21:1539-1558.

32. Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publi-cation bias. Biometrics. 1994;50:1088-1101.

33. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by asimple, graphical test. BMJ. 1997;315:629-634.

34. Duval S, Tweedie R. Trim and fill: a simple funnel-plot-based method of testing andadjusting for publication bias in meta-analysis. Biometrics. 2000;56:455-463.

35. Orwin RG. A fail-safe N for effect size in meta-analysis. J Educ Stat. 1983;8:157-159.36. Capriola-Hall NN, Wieckowski AT, Ollendick TH, White SW. The influence of social

communication impairments on gaze in adolescents with social anxiety disorder. ChildPsychiatry Hum Dev. 2018;49:672-679.


mailto:[email protected]

https://doi.org/10.1016/j.jaac.2019.06.006



37. Dodd HF, Hudson JL, Williams T, Morris T, Lazarus RS, Byrow Y. Anxiety andattentional bias in preschool-aged children: an eyetracking study. J Abnorm Child Psy-chol. 2015;43:1055-1065.

38. Haller SPW, Doherty BR, Duta M, Kadosh KC, Lau JYF, Scerif G. Attention allocationand social worries predict interpretations of peer-related social cues in adolescents. DevCogn Neurosci. 2017;25:105-112.

39. Heathcote LC, Lau JY, Mueller SC, et al. Child attention to pain and pain tolerance aredependent upon anxiety and attention control: an eye-tracking study. Eur J Pain. 2017;21:250-263.

40. Kleberg JL, Hogstrom J, Nord M, Bolte S, Serlachius E, Falck-Ytter T. Autistic traits andsymptoms of social anxiety are differentially related to attention to others’ eyes in socialanxiety disorder. J Autism Dev Disord. 2017;47:3814-3821.

41. Michalska KJ, Machlin L, Moroney E, et al. Anxiety symptoms and children’s eye gazeduring fear learning. J Child Psychol Psychiatry. 2017;58:1276-1286.

42. Price RB, Siegle GJ, Silk JS, et al. Sustained neural alterations in anxious youth per-forming an attentional bias task: a pupilometry study. Depress Anxiety. 2013;30:22-30.

43. Price RB, Rosen D, Siegle GJ, et al. From anxious youth to depressed adolescents:prospective prediction of 2-year depression symptoms via attentional bias measures.J Abnorm Psychol. 2016;125:267-278.

44. Schmidtendorf S, Wiedau S, Asbrand J, Tuschen-Caffier B, Heinrichs N. Attentionalbias in children with social anxiety disorder. Cogn Res Ther. 2018;42:273-288.

45. Seefeldt WL, Kramer M, Tuschen-Caffier B, Heinrichs N. Hypervigilance and avoidance invisual attention in childrenwith social phobia. J BehavTher ExpPsychiatry. 2014;45:105-112.

46. Shechner T, Jarcho JM, Britton JC, Leibenluft E, Pine DS, Nelson EE. Attention bias ofanxious youth during extended exposure of emotional face pairs: an eye-tracking study.Depress Anxiety. 2013;30:14-21.

47. Shechner T, Jarcho JM, Wong S, Leibenluft E, Pine DS, Nelson EE. Threats, rewards,and attention deployment in anxious youth and adults: an eye tracking study. BiolPsychol. 2017;122:121-129.

48. Tsypes A, Owens M, Gibb BE. Suicidal ideation and attentional biases in children: aneye-tracking study. J Affect Disord. 2017;222:133-137.

49. Rosenthal R. The file drawer problem and tolerance for null results. Psychol Bull. 1979;86:638.

50. In-AlbonT,Kossowsky J, Schneider S.Vigilance and avoidance of threat in the eyemovementsof children with separation anxiety disorder. J Abnorm Child Psychol. 2010;38:225-235.

51. In-Albon T, Schneider S. Does the vigilance-avoidance gazing behavior of children withseparation anxiety disorder change after cognitive-behavioral therapy? J Abnorm ChildPsychol. 2012;40:1149-1156.

52. Gamble AL, Rapee RM. The time-course of attentional bias in anxious children andadolescents. J Anxiety Disord. 2009;23:841-847.

53. Chen NT, Thomas LM, Clarke PJ, Hickie IB, Guastella AJ. Hyperscanning andavoidance in social anxiety disorder: the visual scanpath during public speaking. Psy-chiatry Res. 2015;225:667-672.

54. Lazarov A, Abend R, Bar-Haim Y. Social anxiety is related to increased dwell time onsocially threatening faces. J Affect Disord. 2016;193:282-288.

55. Buckner JD, Maner JK, Schmidt NB. Difficulty disengaging attention from social threatin social anxiety. Cogn Ther Res. 2010;34:99-105.

56. Liang CW, Tsai JL, Hsu WY. Sustained visual attention for competing emotional stimuliin social anxiety: an eye tracking study. J Behav Ther Exp Psychiatry. 2017;54:178-185.

57. Schofield CA, Johnson AL, Inhoff AW, Coles ME. Social anxiety and difficulty disen-gaging threat: evidence from eye-tracking. Cogn Emot. 2012;26:300-311.

58. Wieser MJ, Pauli P, Weyers P, Alpers GW, Muhlberger A. Fear of negative evaluationand the hypervigilance-avoidance hypothesis: an eye-tracking study. J Neural Transm(Vienna). 2009;116:717-723.

59. Waters AM, Craske MG. Towards a cognitive-learning formulation of youth anxiety: anarrative review of theory and evidence and implications for treatment. Clin Psychol Rev.2016;50:50-66.


60. Amso D, Scerif G. The attentive brain: insights from developmental cognitive neuro-science. Nat Rev Neurosci. 2015;16:606-619.

61. Paus T. Mapping brain maturation and cognitive development during adolescence.Trends Cogn Sci. 2005;9:60-68.

62. Beesdo K, Knappe S, Pine DS. Anxiety and anxiety disorders in children and adolescents:developmental issues and implications for DSM-V. Psychiatr Clin North Am. 2009;32:483-524.

63. Grafton B, Southworth F, Watkins E, MacLeod C. Stuck in a sad place: biased atten-tional disengagement in rumination. Emotion. 2016;16:63-72.

64. Pergamin-Hight L, Naim R, Bakermans-Kranenburg MJ, van IJzendoorn MH, Bar-Haim Y. Content specificity of attention bias to threat in anxiety disorders: a meta-analysis. Clin Psychol Rev. 2015;35:10-18.

65. Mogg K, Millar N, Bradley BP. Biases in eye movements to threatening facial expressionsin generalized anxiety disorder and depressive disorder. J Abnorm Psychol. 2000;109:695-704.

66. Wermes R, Lincoln TM, Helbig-Lang S. How well can we measure visual attention?Psychometric properties of manual response times and first fixation latencies in a visualsearch paradigm. Cogn Ther Res. 2018;41:588-599.

67. Waechter SN AL, Wright C, Hyatt A, Oakman J. Measuring attentional bias tothreat: reliability of dot probe and eye movement indices. Cogn Ther Res. 2014;38:313-333.

68. Rinck M, Reinecke A, Ellwart T, Heuer K, Becker ES. Speeded detection and increaseddistraction in fear of spiders: evidence from eye movements. J Abnorm Psychol. 2005;114:235-248.

69. Huijding J, Mayer B, Koster EH, Muris P. To look or not to look: an eye movementstudy of hypervigilance during change detection in high and low spider fearful students.Emotion. 2011;11:666-674.

70. Dodd HF, Vogt J, Turkileri N, Notebaert L. Task relevance of emotional informationaffects anxiety-linked attention bias in visual search. Biol Psychol. 2017;122:13-20.

71. Kuckertz JM, Strege MV, Amir N. Intolerance for approach of ambiguity in socialanxiety disorder. Cogn Emot. 2017;31:747-754.

72. Liu H, Li X, Han B, Liu X. Effects of cognitive bias modification on social anxiety: ameta-analysis. PLoS One. 2017;12:e0175107.

73. Mogg K, Waters AM, Bradley BP. Attention Bias Modification (ABM): Review of Effectsof Multisession ABM Training on Anxiety and Threat-Related Attention in High-Anxious Individuals. Clin Psychol Sci. 2017;5(4):698-717.

74. Bar-Haim Y. Research review: attention bias modification (ABM): a novel treatment foranxiety disorders. J Child Psychol Psychiatry. 2010;51:859-870.

75. Lazarov A, Pine DS, Bar-Haim Y. Gaze-contingent music reward therapy for socialanxiety disorder: a randomized controlled trial. Am J Psychiatry. 2017;174:649-656.

76. Linetzky M, Kahn M, Lazarov A, Pine DS, Bar-Haim Y. Contingent music rewardtherapy for clinically anxious 7- to 10-year-olds: an open multiple baseline feasibilitystudy. J Clin Child Adolesc Psychology. 2019 Mar 25;1-8. https://doi.org/10.1080/15374416.2019.1573685 [Epub ahead of print].

77. Cristea IA, Mogoase C, David D, Cuijpers P. Practitioner review: Cognitive biasmodification for mental health problems in children and adolescents: a meta-analysis.J Child Psychol Psychiatry. 2015;56:723-734.

78. Heeren A, Mogoase C, McNally RJ, Schmitz A, Philippot P. Does attention biasmodification improve attentional control? A double-blind randomized experiment withindividuals with social anxiety disorder. J Anxiety Disord. 2015;29:35-42.

79. De Voogd EL, Wiers RW, Prins PJ, Salemink E. Visual search attentional bias modi-fication reduced social phobia in adolescents. J Behav Ther Exp Psychiatry. 2014;45:252-259.

80. Waters AM, Zimmer-Gembeck MJ, Craske MG, Pine DS, Bradley BP, Mogg K. Lookfor good and never give up: a novel attention training treatment for childhood anxietydisorders. Behav Res Ther. 2015;73:111-123.

www.jaacap.org 99

https://doi.org/10.1080/15374416.2019.1573685

https://doi.org/10.1080/15374416.2019.1573685


FIGURE S1 Funnel Plot for Between-Group First-Fixation Analysis

FIGURE S2 Funnel Plot for Between-Group Dwell-Time Analysis

99.e1 www.jaacap.org Journal of the American Academy of Child & Adolescent PsychiatryVolume 59 / Number 1 / January 2020

LISK et al.


Documents

Systematic Review and Meta-Analysis: Eye-Tracking of