Measurement of Target Engagement and Network Analysis of Change Mechanisms in Web-Based Interpretation Bias Training for Anxiety

Anxiety disorders are prevalent but undertreated. Cognitive bias modification (CBM) programs target disorder-relevant processing biases without needing a therapist, providing a potential way to improve access to treatment (e.g., via web- or app-based delivery). Although CBM results have been mixed, overall the evidence suggests that CBM for interpretation biases (CBM-I) may reduce anxiety (and comorbid depression). CBM-I provides repeated practice resolving ambiguous threat-relevant situations in a benign way, aiming to foster more flexible thinking and shift the rigidly negative interpretation style associated with emotional disorders.
However, the psychometric properties of common measures of interpretation biases, the purported mechanism of change in CBM-I, are unacceptable in some studies and unreported in most. Unreliable or invalid measures make it difficult to evaluate target engagement in clinical trials and to assess the mechanism’s effects on symptoms. To advance research on cognitive mechanisms in anxiety, Study 1 evaluated the structural validity (factor structure and internal consistency) of a Recognition Ratings measure used as a primary measure of interpretation biases in CBM-I trials. Using baseline data from a trial run with anxious community adults (N = 749) on a public research website, initial confirmatory factor analysis (CFA) models inferred from the literature were unsupported. Exploratory factor analysis (EFA) suggested three factors (positive threat, negative threat, nonthreat) and potential items to exclude, and exploratory CFA models suggested the need to correlate the errors of items from the same scenario. Exploratory CFA models with (a) the three factors based on a subset of 28 threat and nonthreat items or (b) two factors (positive threat, negative threat) based on all 18 threat items had well-defined factors with generally acceptable internal consistency and construct reliability. Factor determinacy was acceptable for the negative threat and nonthreat factors but mixed for positive threat. Overall CFA model fit was mixed. Exploratory CFA models of the threat items supported the positive threat and negative threat factors as two distinct constructs (i.e., not method artifacts). The study provides initial evidence for the measure’s use in assessing positive and negative biases, although ongoing construct validation will further optimize its use. The two-factor model was selected for Study 2 given its balance of 9 positive threat and 9 negative threat items (unlike the three-factor model with 5 positive threat and 8 negative threat items).
Moreover, most research on interpretation biases and anxiety has conceptualized anxiety symptoms as caused by an underlying disorder, whereas network theory posits that disorder is constituted by causal relations among symptoms themselves, allowing for analysis of anxiety as a complex system. Adopting this perspective, Study 2 uses data from the same randomized trial to test the effects of positive CBM-I (vs. 50-50 CBM-I and no-training comparison conditions) on networks of interpretation biases, individual anxiety symptoms, and related impairment. Cross-sectional network models testing positive CBM-I’s effects on the mean levels of nodes at each of three time points (baseline, Session 3, Session 6) revealed significant direct effects (causal) and indirect effects (possibly causal), suggesting potential pathways via which positive CBM-I deactivates the network. In addition, analyses of each condition’s within-person relations in temporal and contemporaneous networks across the time points revealed descriptively lower connectivity in positive CBM-I than in 50-50 CBM-I and no-training. Although these condition differences in connectivity need to be tested statistically, this tentatively suggests that positive CBM-I may also destabilize the network’s structure, which may reduce the vulnerability of the network to a stable state of high activation. Together, these studies strengthen inferences about interpretation biases and advance understanding of cognitive mechanisms of change in anxiety.