Biased information-processing for threat material is a key component of anxiety, with many data suggesting biases in attention towards threat and in the interpretation of ambiguous (verbal) material (Tables 1,2,3). Prior studies have demonstrated anxiety-based disruptions on attentional mechanisms, and in the interpretation of verbal ambiguous material, such as words and scenarios. However, there are gaps in the literature with fewer studies measuring differences in the evaluation of non-verbal threatening stimuli, such as negative face-emotions, and in the acquisition of fear and avoidance to potential threat cues.
Here, we aimed to follow up some gaps identified in our knowledge of anxiety-based threat biases in children, by exploring differences in the classification of threatening facial expressions and in the acquisition of avoidant responses to a masked threatening face.
Specifically, within our sample of 10-year old twins, we compared those who had been selected for high anxiety symptoms at age 7 and who also met criteria for an anxiety disorder at age 10, with those who did not meet diagnostic criteria and reported lower anxiety scores at age 7 or 10 years. Clear support for one of our hypotheses was found: subjects with an anxiety problem were more likely to avoid a cue paired with a masked angry facial expression, compared to children without anxiety problems. This replicates a prior study using this task in another sample of children . Interestingly in both studies, this tendency appeared to emerge over the course of the task across blocks between children with and without anxiety problems. Results were less clear for another hypothesis: while the ability to recognize angry, fear, and sad facial expressions was not associated with anxiety, children with an anxiety problem were significantly better than those without problems at identifying disgust faces. Contrary to our predictions on the origins of threat biases, no support was found for genetic effects, and all of the variance was explained by non-shared environmental effects, which could include measurement error. This contrasted with the ability to identify disgust faces, which in a previous study of the same sample showed moderate genetic and large environmental contributions , consistent with joint roles for nature and nurture.
These data are subject to various limitations. First, it is possible that the avoidance task is less reliable than the facial recognition task, and that this reduced the twin correlations. This may explain the lack of heritability on threat avoidance, while artificially inflating non-shared environmental contributions which includes measurement error. More particularly, different methodologies for assessing threat biases are likely to have different levels of reliability; an experimental task is likely to be less reliable than a questionnaire measure, resulting in greater non-shared environmental and subsequently smaller genetic influences. Investigating the genetic and environmental origins of such threat biases using experimental tasks rather than questionnaires would therefore benefit from attempts to quantify and improve on the psychometrics of such tasks.
A second set of caveats concerns methods used to ascertain diagnostic status of our ‘anxious’ subjects. While we relied on a version of a well-validated structured interview  -the DAWBA - its’ application of the functional impairment criterion described in DSM, as well as meeting symptom-threshold may have under-identified those with anxiety problems reported in the present sample. Indeed other studies have often reported discrepancies in prevalence rates of anxiety disorders depending on whether clinically-significant distress and impairment are incorporated in the diagnostic procedures [110, 111], with far lower rates reported when ‘clinical impact’ is considered. In the present study, using both clinical impact and symptoms to determine diagnosis, 5% of the sample met criteria for an anxiety disorder. While this figure is not particularly low when considered against other reported prevalence rates of this age range, which can vary between 2.6% to 41.2% , it is lower than expected given that most children in the sample were selected for high anxiety. This allows the possibility that many more children in our sample had high anxiety, but only some also experienced distress and impairment. Indeed as a group, non-diagnosed children in the ECHO sample reported a mean anxiety score on the SCARED that approached clinical cut-off for that measure. As we were interested in assessing anxiety-based differences in threat biases, it was even more important for us to compare children who met clinical diagnosis with children who not only did not meet clinical diagnosis, but who also reported fewer anxiety problems at age 7 and at age 10. Thus, we selected our original control twin pairs as the comparison subjects for this analysis (rather than simply those who did not meet anxiety disorder at age 10 years). A second issue relates to the use of parent-reported data to generate computerized algorithms to detect children meeting criteria for an anxiety disorder. This approach raises issues on the accuracy of parents as informants, in the absence of clinical interview. Future research should employ multi-method, multi-informant measures to increase the validity of reported associations between anxiety and information-processing biases.
A final issue concerns heterogeneity in diagnostic subtypes and co-morbid conditions within the group meeting criteria for at least one current anxiety disorder. Although collapsing across subtypes is sub-optimal, adequate numbers precluded examination of more specific links with biases in information-processing. Nevertheless studies of anxiety subtypes in children typically yield strong cross-sectional and longitudinal comorbidity; similar mental health histories; and large overlap in genetic liability [112, 113], lending justification to analyzing anxiety disorders as a single group in the first instance.
Despite these limitations, our data offer some interesting extensions into cognitive phenotypes of child anxiety. One key finding relates to the association between anxiety and avoidance. Prior studies examining avoidance in anxiety have used questionnaires or observational measures that do not correspond closely to definitions of established theories . According to such theories, through associative learning, a neutral conditioned stimulus (e.g. color card, CS+) acquires fear-eliciting properties of an aversive unconditioned stimulus (e.g. angry face, UCS). Avoidance is then employed and reinforced through its fear-reducing capacity. The paradigm in the current study was designed to model the acquisition of avoidance learning. Both the current and a prior study using this paradigm  show that children with anxiety problems, defined by questionnaire and diagnostic measures, are more likely than children with fewer anxiety problems to ‘avoid’ a cue associated with a threatening face in favor of a cue associated with a neutral face. These data fit in well with suggestions that anxiety in childhood can be characterised by a pattern of initial vigilance (as suggested by visual probe studies showing attention-orienting towards threats) followed by subsequent avoidance in response-selection. Interestingly, this vigilance-avoidance pattern has also been described in adults with anxiety . These findings are not too surprising when placed in the context of clinical features of anxiety disorders in children and adults, which often involve marked fear and avoidance of the feared object in tandem. Indeed avoidant strategies are thought to maintain the marked fear. Finally, using model-fitting estimates, the current study does not support the role of inherited factors in shaping these avoidant behaviours but instead points to the importance of non-shared environmental variance. Future studies should try to assess the contribution of specific, measured environmental influences that account for this source of variance.
A second interesting finding was the association between anxiety and biased recognition of disgust faces, a bias that we previously reported to be influenced by genetic factors . This bias in the recognition of disgust among children with anxiety problems occurred in the context of similar abilities to identify angry, fear and sad expressions relative to children without anxiety. There is some support from pediatric samples corroborating the relationship between trait anxiety and disgust sensitivity . Preliminary data also finds greater sensitivity to disgust in adults with high levels of social anxiety, relative to adults with lower levels. This sensitivity is manifest through behavioral ratings and reaction times to disgust faces as well as in patterns of brain activation . Why would anxious individuals be more sensitive to disgust stimuli? Given that biases associated with threat are a characteristic of children with anxiety, disgust could reflect a social threat (e.g. rejection). Disgust could also signal a physical threat (e.g. contamination). Further work is needed to clarify the role of disgust in children’s anxiety problems. It would also be interesting to explore whether disgust faces similarly affect other stages of information-processing, such as by capturing or interfering with attentional resources.
Increased avoidance of feared stimuli and sensitivity to disgust stimuli may contribute to pathological anxiety in the long-term by maintaining anxious states. Thus therapeutic interventions that aim to extinguish acquired fear and avoidant associations, or modify biases in the processing of threats may be particularly helpful in combating anxiety. Exposure based interventions, which capitalize on fear reduction through extinction learning have been used effectively to treat adult anxiety and to some extent in children with anxiety problems. As fear to the CS + declines via extinction, avoidance will no longer be needed to attenuate fear. Alternatively, children could be taught counter strategies to terminate conditioned fear, such as the use of approach-based strategies. For disgust sensitivity, new bias modification programs could be developed and implemented to train children to re-evaluate initial impressions based on other evidence, such as the presence of positive emotional expressions. While cognitive bias modification training tasks appear effective in manipulating attention away from threats, or leading to the re-appraisal of ambiguous scenarios, these training tasks have not yet been extended to manipulate the labelling of ambiguous non-verbal cues, such as face-emotions. If children with high anxiety do show greater recognition of disgust faces, this could be a new target for such computerized training paradigms - with the aim to reduce negative perceptions and thus anxious mood-states.