Cognitive models of panic disorder (PD; Clark
1986; Margraf & Schneider,
1990) postulate that negative interpretation biases (IBs), namely the tendency to interpret ambiguous, body-related information in a negative or catastrophizing manner, are central to the development and maintenance of the disorder. IBs in the context of PD typically relate to catastrophic misinterpretations of ambiguous bodily symptoms occurring in an arbitrary everyday-life situation, for example interpreting sudden palpitations as a sign of a heart-attack. Evidence for the central role of body-related IBs in the context of panic comes from several lines of research. For example, it has been shown that IBs are strongly associated with anxiety sensitivity (Olthuis et al.,
2012; Teachman,
2005; Zahler et al.,
2020), a concept describing the concern about experiencing symptoms of anxiety which in turn is predictive of the new onset of PD (Woud et al.,
2014) and commonly applied as a subclinical panic analog (cf. Zahler et al.,
2020). In clinical samples, Teachman et al. (
2007) found that compared to healthy controls, patients with PD interpreted ambiguous scenarios describing bodily symptoms as more threatening (for similar results, see McNally & Foa
1987; Richards et al.,
2001), and reported greater anxiety levels during a symptom provocation, with anxiety levels in turn being associated with the IBs. In another cross-sectional study Hermans et al. (
2010) found that patients with PD showed more negative IBs concerning bodily symptoms compared to patients with other anxiety disorders and healthy controls. Besides this cross-sectional evidence, it has been shown that IBs were predictive of new onset of PD (Woud et al.,
2014). Moreover, the reduction of IBs precedes symptom improvement during cognitive behavioral therapy for PD but not vice versa (Teachman et al.,
2010). This provides further evidence for IBs not only being a correlate but also a risk factor of PD (for reviews on the association between IBs and anxiety disorders, see Hirsch et al.,
2016; Ohst & Tuschen-Caffier,
2018). While these findings offer correlational evidence for the central role of IBs in panic, investigating the potential causal role of IBs is the next crucial and critical step to investigate predictions of cognitive models of PD. This step, in turn, requires manipulating IBs, i.e., reducing and inducing them, followed by examining the effects of such a manipulation on panic-related symptoms (Kraemer et al.,
1997).
Techniques for the manipulation of IBs have been developed within the framework of Cognitive Bias Modification for Interpretations (CBM-I; Koster et al.,
2009; Woud & Becker,
2014). CBM-I involves a range of computerized tasks during which participants see ambiguous, incomplete information (e.g., an open-ended sentence) that can only be completed according to participants’ allocated training condition, e.g., in a positive or negative manner. One example are sentence completion paradigms (cf. MacLeod & Mathews,
2012) during which participants are presented with disorder-related, ambiguous open-ended sentences (e.g.,
A strange feeling in my stomach signals that I am very…), and are instructed to resolve this ambiguity by finishing a subsequently presented word fragment (e.g., positive resolution: “h_ppy”,
happy; negative resolution: “i_l”
ill). Importantly, each word fragment is chosen such that it resolves the ambiguity in a training-congruent manner. To illustrate, in a training designed to induce more positive interpretations, positive and functional words are used for the word fragment, and the ambiguity is thus consistently resolved positively. To date, there are several studies showing positive effects of CBM-I on (analog) symptoms of various psychopathologies (see for example Hirsch et al.,
2018; Woud et al.,
2021; for a recent network meta-analysis, see Fodor et al.,
2020), yet the literature in the context of PD is still in its infancy.
In one of the first studies, Steinman & Teachman (
2010) tested the effects of a positive versus control CBM-I training on panic-related IBs, fear of bodily symptoms, and responses to a panic symptom provocation task in individuals scoring high on anxiety sensitivity. As expected, the positive training led to a more positive IB and reduced anxiety sensitivity, compared to the control conditions. However, at post-training, the groups did not differ in more panic-specific symptomatology, including fear of body symptoms or fear and avoidance during the symptom provocation task (for similar results, see Clerkin et al.,
2015; MacDonald et al.,
2013). Capron et al. (
2017) combined CBM-I with a psychoeducation intervention and reported more promising results—they also successfully induced a more positive IB via a positive compared to a control condition receiving only psychoeducation and additionally found that positively trained participants reported both less fear during a symptom provocation task and lower scores on anxiety sensitivity. Finally, a recent study investigated the effects of a CBM-I training in a sample including diagnosed PD patients. Findings were promising such that a more positive IB was induced, and there was a significant panic symptom reduction from pre- to post-intervention (Beard et al.,
2016). However, since the study did not include a control condition these results have to be interpreted with caution.
Discussion
The aim of the present study was to investigate whether a positive versus a negative panic-related CBM-I training would induce training-congruent IBs in comparison to a neutral control condition. The second aim was to test whether these effects would generalize to another, far transfer measure of IBs, namely the SST. The third aim was to investigate whether the positive and negative CBM-I training would have corresponding effects on participants’ reactions to various panic-related symptom provocation tasks, and thus whether panic-related IBs might be causally involved in the exacerbation of symptoms.
With regard to our first aim, our results were mixed in that we found a successful induction of a positive IB on the ERT in comparison to the negative and the control condition. However, the induction of a negative IB was only partly successful as the negative condition did not significantly differ from the control condition. Concerning our second aim, results indicated only a limited generalization effect on another measure of IBs, the SST. Here, we found a more positive IB in the positive than in the negative condition, yet neither the positive nor the negative condition differed significantly from the control condition. Finally, regarding our third aim, we found no effects of CBM-I on the symptom provocation tasks as indicated by no significant group differences regarding the increase in panic-related symptoms in the second symptom provocation compared to baseline.
Results concerning our first aim are generally in line with previous research on the effects of CBM-I in manipulating interpretations of bodily sensations in the context of panic (e.g., Capron et al.,
2017; Steinman & Teachman,
2010). Comparable to previous work, we found that the positive condition successfully induced a more positive IB on the ERT compared to a neutral control and negative condition, whereby the latter comparison extends previous findings. However, as the negative condition did not differ from the neutral control condition, our aim of inducing a negative IB was not met. Hence, the hypotheses concerning the potentially causal role of IBs as suggested by cognitive models of panic could only be partly investigated. One reason for this finding could be that the strongly negative interpretations (e.g., sudden palpitations as a sign of a heart attack) were too extreme to be perceived as plausible in our healthy sample, and thus had no impact on participants’ interpretational style. Supporting this hypothesis, participants reported only little anxiety in response to the bogus feedback concerning their heart rate, questioning whether participants were actually susceptible to the induction of negative cognitions concerning their body symptoms.
Regarding the second aim of our study, our findings indicate a limited generalization of panic-related CBM-I effects to a more distal measure of IBs, the SST, which mirrors the findings by Steinman & Teachman (
2010). The most parsimonious explanation for these findings relates to the small to medium effect sizes found for the post-training group comparisons on the ERT (see also Salemink et al.,
2022). Since the ERT is closely matched with the CBM-I training, i.e., its stimuli and operationalization, one would expect medium-to-large effects on this measure (for a recent meta-analysis, see Martinelli et al.,
2022). However, since we did not find such effects, it seems unlikely to then find large effects on the SST, a task that is conceptually even further away from the training. A second potential explanation for our mixed results is that we used a heterogeneous set of stimuli for both the training and the IBs’ assessment (i.e., ERT and SST). However, not all stimuli and related cognitions may have been idiosyncratically relevant for all participants (cf. Schneider & Schulte,
2007). An aim for future studies might therefore be to redesign all stimuli such that they have a greater overall impact on and match with participants’ IBs, e.g., by targeting the subtypes often reported in PD (see Sansone & Sansone
2009). For instance, patients with PD often report symptoms concerning one particular system e.g., cardiac, respiratory, gastrointestinal, or vestibular. Identifying the bodily symptom that is most relevant and then matching the CBM-I and assessment stimuli to it might therefore increase the training’s internal validity and ultimately generalization.
Finally, our results concerning the third aim suggested that the induced variation in panic-related IBs did not affect participants’ reaction to the symptom provocation tasks. One interpretation of this result is that IBs might not be causally involved in the exacerbation or reduction of panic-related symptoms. This interpretation, however, would be in contrast with earlier findings in clinical populations where IBs have been shown to be predictive of panic disorder onset (Woud et al.,
2014) and their reduction predicted symptom reduction during treatment (Teachman et al.,
2010). In this light, we consider a more likely explanation to be that the variance in IBs induced by our CBM-I conditions was insufficient to have effects on panic-related symptoms in our sample. In this case we would not expect an effect of the CBM-I on symptoms (cf. Clarke et al.,
2014). Notably, we were unable to shift IBs in a negative direction, and it seems plausible that we would need to induce a range of IBs from negative to benign or positive rather than only from relatively benign to positive to observe meaningful effects on panic symptom provocation tasks in an analog sample.
Another potential explanation for our null findings, also put forward by Steinman & Teachman (
2010), is that the symptom provocation might not have been challenging enough to provide variance in participants’ responses, and thus may not have been an ideal panic analog. Although we aimed to increase the stressfulness of the symptom provocation tasks by increasing the number of different tasks compared to Steinman & Teachman (
2010), our exploratory results showed that participants indeed reported an increase in body-related symptoms during the symptom provocation tasks, yet no increase of anxiety. Our findings could therefore indicate that in our analog sample a more impactful stressor would have been necessary for an effect of a positive versus (in our sample) relatively neutral IB to unfold. One option for further research is therefore to include symptom provocation tasks that have been shown to produce fear in healthy participants more reliably such as a vital CO2 challenge as applied by Capron et al. (
2017). Another consideration regarding our stressor concerns idiosyncrasy. While we aimed to induce a variety of symptoms, it might be necessary to assess which symptoms are idiosyncratically relevant to participants to allow valenced interpretations concerning these symptoms to unfold (cf. Schneider & Schulte,
2007). As these interpretations attribute the lack of evidence for a causal role of IBs in panic to limited effectiveness of CBM-I, broader methodological considerations related to the study design that might limit effectiveness need to be taken into account. First, a one session “dose” of CBM-I may not have been sufficient to affect more distal measures of panic-related cognitive processing (for discussions on dose-response effects, see Hallion & Ruscio
2011; Jones & Sharpe,
2017). Similar mixed results on SST have occurred in other studies with single session CBM-I (e.g., Holmes et al.,
2009; Yiend et al.,
2014), while studies with multiple sessions of CBM-I have led to considerable change on the SST (e.g., Hirsch et al.,
2020). Considering our exploratory finding that the SST but not the ERT was associated with the analog symptom measure, a generalization to the SST might be particularly important, as it has been suggested that the interpretational processing assessed via the SST is closely related to symptoms of psychopathology (Würtz et al.,
2022). Accordingly, future research could investigate multiple CBM-I training sessions to increase generalization effects. However, another reason for the differential effects of CBM-I on the ERT and the SST might relate to the suggestion that interpretational processing consists of both automatic and reflective processing (e.g., Hirsch et al.,
2016), and CBM-I might have a stronger effect on either of those. The questions, to which extent CBM-I targets rather automatic or reflective processing (e.g., Bowler et al.,
2012) and which of these aspects is assessed via the ERT and particularly through the SST (Würtz et al.,
2022) are unresolved and remain important subjects for discussion in future research. For example, applying measures of IBs that limit reflective processing in CBM-I studies, such as neurophysiological correlates (e.g., Feng et al.,
2019), together with measures like the SST might shed light on both questions, which targets are aimed at more strongly via CBM-I and which processes are reflected through each measure.
Another factor that could potentially affect generalization is the context in which CBM-I is embedded. While our study investigated the effects of CBM-I as a stand-alone procedure, Capron et al. (
2017) compared CBM-I in combination with psychoeducation to psychoeducation only. Their results showed that including CBM-I significantly reduced participants’ anxiety sensitivity, indicating successful generalization to another risk factor relevant to panic. Providing participants with a rationale on the relationship between IBs and symptoms in future studies or instructing them to actively apply the interpretations presented in the training in their daily life might therefore aid in improving generalization to symptoms. This approach might be useful in maximizing potential effects of CBM-I as is desired in a clinical setting to reduce symptoms. However, its application in proof-of-concept studies like the present one might be limited due to a potential increase in demand effects when participants know the rationale.
Another consideration regarding the study setup arises when comparing our study to Capron et al. (
2017) and concerns the population from which our sample was drawn. While Capron et al. (
2017) conducted their study with a sample with elevated anxiety sensitivity, our sample consisted of participants with a moderate level of fear of body symptoms as measured via the BSQ. Our aim in deviating from previous research that recruited based on the Anxiety Sensitivity Index was to specifically recruit participants who reported a medium level of fear of bodily symptoms instead of general anxiety sensitivity, as we considered this fear to be more specific to panic and thus our training. This was based on evidence that only anxiety sensitivity in relation to body symptoms was specifically associated with panic, while other dimensions (i.e., social or cognitive concerns) were related to broader psychopathology (Olthuis et al.,
2014). However, our exploratory results that the BSQ was neither associated with the ERT nor with panic-related symptoms at pre-training question the relevance of this measure in the current study context. Additionally, as a side-effect of our recruitment process, participants’ overall anxiety sensitivity was considerably lower than in previous studies (
M = 13.09 in our sample compared to
M = 28.57 in Capron et al.
2017) which may have resulted in a sample less susceptible to the general set-up of this study and the CBM-I training in particular. That is, they may not have started with sufficiently negative IBs for positive training to meaningfully change reactions to the symptom provocation task, and the negative training may not have been sufficiently plausible to alter IBs in a more negative direction.
Our findings need to be interpreted in light of several limitations. First, the repeated application of symptom provocation tasks might have results in practice and habituation effects that might have overshadowed potential effects of CBM-I on the symptom provocation. Further, the fixed order of symptom provocation tasks did not allow to investigate order effects which would have been possible through a counterbalanced design. Second, our cover story that divided the training and the symptom provocation tasks into two different studies might have impeded participants in applying the interpretations from CBM-I to the body symptoms experienced during the provocation tasks. This, in turn, may have reduced generalization, and seems a sub-optimal approach when studying the effects of a training that in fact requires the active application of a newly trained cognitive style. As CBM-I in combination with psychoeducation resulted in more promising findings (Capron et al.,
2017), future studies might therefore want to provide participants with a rationale for the interplay of IBs and (analog) symptoms and encourage them to actively apply those interpretations for example during symptom provocation. Third, our sample mostly consisted of healthy female participants with at least A-level education and German as a native language. Our results are thus potentially not generalizable to more diverse populations and are not translatable to patient populations. A fourth consideration concerns our control condition. Our aim was to apply a neutral control condition by applying a training with scenarios not related to the interpretation of bodily sensations. However, upon reflection, our control training may not have actually been completely
neutral. That is, although trials were not related to (and thus neutral) with respect to panic, they were disambiguated in either a positive or negative manner, and thus not neutral in terms of valence. As discussed earlier, in our rather healthy sample, the positive endings may have been more relevant to the participants than the negative endings, rendering the control training mildly positive and thus dampening the effect size for the difference between the positive and the negative condition. Future research with the aim to investigate effects of CBM-I in comparison to a neutral control condition should therefore apply truly neutral stimuli, e.g., neutral descriptions of everyday situations as a control condition (for a discussion of adequate control conditions in the context of CBM, see Blackwell et al.,
2017).
To conclude, our study provided further evidence that CBM-I can induce a positive IB concerning bodily symptoms in a healthy sample on an IB measure similar to the CBM-I training, the ERT. However, our results on the induction of a negative IB were mixed delivering no evidence as to their potential causal role in the etiology of panic. Further, due to our finding that CBM-I effects did not fully generalize to a more distal measure of IBs, the SST, and had no effects on a panic-related symptom provocation, future research on improving generalization for the purpose of experimental tests of causality, for example via adaptations of the CBM-I training and the study design, is clearly warranted.