Introduction

Threat perception refers to how an individual detects, attends, interprets and responds to environmental stimuli that may pose risk or danger [1]. Threat perception is therefore an evolutionarily important capacity, fundamental to survival and adaptive functioning [2]. Reduced capacity to accurately and rapidly detect threat in one’s immediate environment has direct implications for personal safety and wellbeing. In a social context, it is vital to be sensitive to the information conveyed in facial expressions to assess whether a specific individual is a source of potential harm or danger. In a non-social context, it is critical to be able to quickly and accurately infer objective risk from fearful and threatening scenes such as attacks, explosions and mutilations. Conversely, a tendency to overestimate threat in either of these contexts could result in an inefficient state of constant anxiety and hypervigilance [3].

To date, literature on threat perception and mental health has involved two major areas of inquiry; attention to threat-relevant stimuli and interpretation of threat-relevant stimuli. Indeed, in their cognitive model, Hirsch and Mathews [4] argue that habitual biases in attention and interpretation favouring threat content are amongst the most important involuntary (bottom-up) contributing processes to pathological worry. With respect to attentional mechanisms, various disruptions in the presence of threat-relevant stimuli have been observed in both clinical cohorts [2,5] as well as individuals that present with sub-clinical symptoms of psychopathology [6,7]. Such studies have identified differing patterns of attentional disruption, including an attentional bias away from threat (‘avoidance’) and a vigilance-avoidance pattern [8]. However, the most commonly observed finding is an attentional bias towards threat-relevant stimuli (i.e., ‘hypervigilance’), which does not appear to reflect a bias towards negative stimuli more broadly [9].

Based on findings such as these, it has been proposed that an attentional bias towards threat may reflect a core, transdiagnostic factor, linked to both initial onset and subsequent maintenance of mental illness. Thus, it has been argued that an individual who selectively attends to threatening stimuli will live in constant hypervigilance, which leads to maladaptive cognitive, attentional and behavioural patterns (e.g. avoidance, catastrophising, social withdrawal) that each contribute to the development of psychopathology [10]. Any ongoing tendency to selectively attend to the most threatening aspects of one’s environment will then continually reinforce this perception of ever-present danger [1,11,12,13].

As noted, there have also been a number of studies focused on how the interpretation of threat-relevant stimuli is affected by psychopathology. Threat interpretation refers to how readily an individual attributes threatening intentions or characteristics to environmental stimuli [14], and cognitive models of anxiety propose that the tendency to interpret ambiguous information as threatening also plays a causal role in the etiology and maintenance of pathological anxiety [15]. In this literature, most studies have focused on better understanding the hostile attribution bias, whereby others’ ambiguous behaviours are interpreted as being motivated by hostility [16]. For instance, in the ‘ambiguous scenario’ methodology, participants are asked to imagine themselves as the protagonist of a scenario (usually social) in which the threat posed is open to interpretation. Participants’ interpretations of such scenarios are used to index how frequently and readily they perceive hostility (i.e. threat) in their environment. Studies in this literature have identified associations between anxiety and depression and a bias towards quick, frequent and intense interpretations of threat [17,18,19,20]. Another approach that has often been taken has been to present participants auditorily with ambiguous homophones (e.g. die versus dye). Individuals with mental health problems have been found to write the threatening spelling of the word more often than controls [13,21].

Taken together, evidence indicates that both initial attention to threat-related cues, as well as interpretation of ambiguous threat-relevant stimuli, are affected by psychopathology, and reflecting the importance attributed to these biases, considerable focus has been placed on the development of clinical interventions in which these biases are targeted directly [22,23]. However, it remains poorly understood whether there are also abnormalities in the interpretation of stimuli for which there are unambiguous threat cues. Indeed, we are aware of only one study to date involving individuals with mental health problems that has assessed whether threat interpretation abnormalities are seen in relation to such stimuli using validated experimental methods [24]. In this study, participants with schizophrenia and non-clinical controls were shown pictures normatively rated as being either high or low in danger level. Half of the pictures depicted situations (e.g. weather, animals, sporting activities), while the other half showed people’s faces. Although both groups were equally able to differentiate between low and high threat stimuli, people with schizophrenia showed a general tendency to perceive situational stimuli as higher in potential danger, with this bias significantly related to level of positive symptomatology.

These data suggest that the negative interpretation bias associated with psychopathology may not be specific to ambiguous stimuli. However, an important consideration is that the participants in Henry et al.’s [24] study were all diagnosed with a severe mental illness (schizophrenia), and had a long history of being unwell (on average 23 years since initial diagnosis). Confounding social factors associated with chronic mental illness such as unemployment, unstable accommodation and social isolation rather than mental illness may therefore have accounted for the negative interpretation bias to unambiguous stimuli. Examining this capacity in adolescents whose illness is of relatively recent onset is therefore needed to gain greater insight into the nature and specificity of threat perception abnormalities associated with mental illness. Consequently, the aim of the present study was to identify whether threat interpretation abnormalities are also evident in a mixed-clinical cohort of adolescents using Henry et al.’s [24] stimuli. Adolescence represents a critical developmental period in which most adult mental health problems first emerge, and there is now considerable literature showing that adolescents with mental health problems are similar to their adult counterparts in showing a tendency to interpret ambiguous information as threatening [23]. However, potentially important differences between adolescents and adults with mental health problems have also been identified, such as in the patterns of neural activation shown when appraising threat [25]. The present study will provide important clarification on whether the abnormalities interpreting threat in unambiguous stimuli reflect a relatively early, potentially transdiagnostic feature of mental illness, or whether they are instead specifically linked to schizophrenia and/or the adverse social experiences of those with chronic mental illness.

Method

Results

Threat Interpretations

Fig 1A and 1B show danger ratings for the clinical and control participants. The Faces data (see Fig 1A) were analyzed with a 2 (group: clinical, control) x 2 (danger level: high, low) analysis of variance (ANOVA), with danger ratings as the dependent variable. These analyses indicated a main effect of danger level, F(1,82) = 487.71, p < .001, η_p² = .86, which reflected higher danger ratings on normatively rated high danger than low danger faces. However, there was no main effect of group, F(1, 82) = 0.38, p = .541, η_p² < .01, and no interaction between danger level and group, F(1, 82) = 0.20, p = .66, η_p² < .01. Thus, as can be seen in Fig 1A, danger ratings for the clinical and control groups were very closely matched, both for high danger faces (M = 4.88, SD = 0.96, and M = 4.73, SD = 0.98, respectively), and for low danger faces (M = 1.93, SD = 0.76, and M = 1.90, SD = 0.89, respectively).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Danger scores (and standard errors) for (A) high- and low-danger faces, and (B) high- and low-danger situations for controls and participants with mental health problems (minimum and maximum scores are 1 and 7, respectively).

https://doi.org/10.1371/journal.pone.0127167.g001

The Situations data (see Fig 1B) were similarly analyzed with a 2 (group: clinical, control) x 2 (danger level: high, low) ANOVA, with danger ratings as the dependent variable. Again, there was a main effect of danger level, F(1,82) = 651.2, p < .001, η_p² = .89, which reflected higher danger ratings on normatively rated high relative to low danger situations. There was also no main effect of group, F(1, 82) = 0.27, p = .608, η_p² < .01, and no interaction between danger level and group, F(1, 82) = 1.41, p = .239, η_p² = .02. As can be seen in Fig 1B, danger ratings for the clinical and control groups were very closely matched, both for high danger situations (M = 5.22, SD = 0.98, and M = 5.43, SD = 0.77, respectively), and for low danger situations (M = 2.49, SD = 0.76, and M = 2.43, SD = 0.73, respectively).

Correlates of Danger Ratings

Threat interpretation data were then examined in terms of their correlates with cognitive function, mental health and satisfaction with life. Because both groups were equally able to differentiate between high and low threat exemplars, for these analyses we computed a total danger rating score separately for faces and situations. These correlations are reported in Table 1, separately for the two groups. Because a relatively large number of correlations are reported, a Bonferonni correction was applied. To avoid inflating Type II errors of interpretation, the appropriate way to apply this correction is to adjust the alpha level for each comparison not only in relation to the number of comparisons made (i.e. 16), but also in relation to the inter-correlation between the four dependent variables (average intercorrelation between the four dependent variables was .64). This procedure yields a corrected p value of .018. Applying this corrected value, in the clinical group, threat ratings were unrelated to all dependent measures. However, in the control group, two significant correlations emerged. Specifically, higher threat ratings to face stimuli were significantly correlated with both measures of psychopathology (RCMAS, p = .003; RADS, p = .001). According to Cohen’s [33] criteria, these correlations were both moderate to large in magnitude. It can also be seen from the scatterplots depicting these associations in Fig 2A and 2B that there is no evidence that either of these correlations are artefactual.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Correlates of threat attributions.

https://doi.org/10.1371/journal.pone.0127167.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Scatterplots depicting associations between total danger scores for faces and (A) scores on the RCMAS, and (B) scores on the RADS.

https://doi.org/10.1371/journal.pone.0127167.g002

Discussion

The present study shows for the first time that adolescents with mental health problems do not differ from controls in their ability to detect unambiguous threat from either the facial expressions of others or from non-social situations. Thus, no group differences emerged with respect to either overall levels of danger perceived, or in the ability to differentiate normatively judged high- and low-danger faces, or high- and low-danger situations. These data therefore deviate from the only study to date that has assessed this type of threat perception in an adult psychiatric cohort. As noted, Henry et al. [24] found that although individuals with schizophrenia did not differ from controls in their ability to differentiate high- and low-threat faces or situations, they provided higher threat ratings overall to situational stimuli. Specifically, they exhibited a general bias to perceive situational stimuli as being higher in potential danger. The present data indicate that these negative interpretational tendencies are not an early feature of mental illness, but may instead be specifically linked to a diagnosis of schizophrenia and/or the adverse social consequences of a chronic mental disorder. Consistent with the former of these possibilities, Henry et al. [24] identified a significant association between positive symptoms and a negative threat interpretation bias, suggesting a potential specific schizophrenia-related abnormality.

These data also deviate from prior studies in the threat interpretation literature that have investigated responses to ambiguous stimuli. As noted, with few exceptions these studies have shown that psychopathology—both clinical and subclinical—is related to quicker, more frequent and more intense interpretations of threat [13,18,19,20,21,22]. The present results raise the interesting possibility that in adolescence at least, this bias may reflect an ambiguity-dependent effect, and consequently may be neither as pervasive nor problematic as previously thought. Such a possibility is particularly interesting in light of the fact that studies that have identified attentional disruptions have most often used stimuli that is unambiguous with respect to potential threat, as was the case in the present study. The finding of abnormalities in attention—but not interpretation—of unambiguous threat cues therefore indicates that these different components of threat perception may be dissociable.

Given the critical importance of being able to make appropriate interpretations of threatening and non-threatening stimuli in our everyday environment, the finding that mental health problems in adolescence does not affect this capacity is potentially clinically significant. As noted earlier, both hyper-and hypo-vigilance to threat have been linked to negative functional outcomes in everyday life, and indeed the second major finding to emerge in the present study was that higher ratings of threat to the facial stimuli were associated with poorer mental health in the control group. Thus, control participants who reported seeing greater threat in facial stimuli were also those who reported the highest levels of anxiety and depression. The absence of these associations in the clinical group however, perhaps suggests that once a clinically significant level of psychopathology is reached, there is no further tendency to appraise others as more threatening i.e. a ceiling effect is reached. Alternatively, the correlations identified in the control group might reflect associations between threat interpretation and individual differences in personality traits related to anxiety or depression proneness, not state anxiety or depression per se. Consistent with such a possibility, Doty et al. [34] showed that sensitivity to detecting unambiguously threatening faces varied parametrically across the healthy population, and was associated with anxiety-related traits, but not situational fluctuations in anxiety. Finally, it is also important to note that the identification of distinct functional correlates for facial and situational threat also perception aligns with other studies that have also shown that there are meaningful differences between the ability to perceive threat from facial relative to non-facial cues [24,30,32].

Acknowledgments

Thanks go to all the participants who gave their time to participate in our research. We would also like to thank J. Bruno Debruille for helpful comments on an earlier draft of this paper.