ارزیابی تغییرات در تعصبات قضاوت به عنوان مکانیسم درمان شناختی رفتاری برای اختلال اضطراب اجتماعی

Abstract Reductions in judgmental biases concerning the cost and probability of negative social events are presumed to be mechanisms of treatment for SAD. Methodological limitations of extant studies, however, leave open the possibility that, instead of causing symptom relief, reductions in judgmental biases are correlates or consequences of it. The present study evaluated changes in judgmental biases as mechanisms explaining the efficacy of CBT for SAD. Participants were 86 individuals who met DSM-IV-TR criteria for a primary diagnosis of SAD, participated in one of two treatment outcome studies of CBT for SAD, and completed measures of judgmental (i.e., cost and probability) biases and social anxiety at pre-, mid-, and posttreatment. Treated participants had significantly greater reductions in judgmental biases than not-treated participants; pre-to-post changes in cost and probability biases statistically mediated treatment outcome; and probability bias at midtreatment was a significant predictor of treatment outcome, even when modeled with a plausible rival mediator, working alliance. Contrary to hypotheses, cost bias at midtreatment was not a significant predictor of treatment outcome. Results suggest that reduction in probability bias is a mechanism by which CBT for SAD exerts its effects.

2. Results For Study 1, a series of ANOVAs and chi-square tests showed no differences in the variables of interest at pretreatment (BFNE, OPQ, OCQ) across the VRE, EGT, and WL conditions (p's = .213 to .830) or demographic characteristics (SAD subtype, gender, ethnicity, educational achievement, income, relationship status; p's = .402 to .841). Thus, random assignment successfully created three conditions that were comparable at pretreatment with regard to symptom severity, judgmental biases, and demographic factors. Participants receiving EGT reported slightly higher first-exposure SUDS ratings than participants receiving VRE, but this difference was not statistically significant (MEGT = 7.4; MVRE = 6.2; t(40) = −1.817, p = .077). At posttreatment, there were no differences between the EGT and VRE groups on any measure (p's = .348 to .802). Thus participants in the EGT and VRE groups were combined, forming a total of two experimental groups (Treated [EGT + VRE], Not Treated [WL]). Independent samples t-tests and chi-square tests were then conducted to determine whether participants from the uncontrolled trial (Study 2) were significantly different in terms of symptom severity, judgmental biases, or demographics at the pretreatment assessment from participants in the controlled trial (Study 1). There were no significant differences between Study 1 and Study 2 on any of the metrics listed above (p's = .254 to .969); as such, participants from Study 2 were added to the Treated group from Study 1 to increase sample size and statistical power. With regard to SAD subtype, there were no significant between-group differences in cost bias or probability bias at pretreatment or working alliance at Session 1 (p's = .064 to .691); however, participants with generalized SAD reported significantly higher levels of social-evaluative fears at pretreatment than participants with non-generalized SAD (Mgeneralized = 44.69; Mnon-generalized = 36.80; t(74) = −3.362, p = .001). To determine whether or not symptom improvement was statistically mediated by reductions in cost bias and probability bias (Criterion 1) and whether CBT caused threat reappraisal (Criterion 2), a multiple mediators path model was tested. Residualized gain scores were first computed using data from pretreatment and posttreatment to represent a measure of change in social anxiety symptoms (BFNE) and threat appraisal (OPQ, OCQ) during treatment. Residualized gain scores control for initial symptom severity and measurement error associated with repeated assessment and thus have advantages over other measures of change (Steketee & Chambless, 1992). Residualized gain scores were calculated by subtracting the standardized pretreatment scores, which were multiplied by the correlation between the standardized scores at pretreatment and posttreatment, from the posttreatment scores. Using this formula, lower residualized gain scores reflect greater reductions in symptoms. Next, a model was tested that included paths for 1) the effect of treatment on the mediators (pre-to-post changes in probability bias and cost bias) (i.e., Criterion 2); 2) the effect of the mediators on treatment outcome (pre-to-post changes in social anxiety symptoms), 3) correlations between the two mediators, and 4) the indirect effect of treatment on treatment outcome through pre-to-post changes in probability bias and cost bias (i.e., Criterion 1). The multiple mediators path model is presented in Fig. 3. Overall, the model fit the data well: Model χ2(1) = 1.652, p = .199; RMSEA = .081 [.000, .294]; CFI = .995; TLI = .967; SRMR = .024, with the exception of the upper limit of the RMSEA confidence interval. First, as predicted, treatment had a significant effect on threat appraisals; receiving treatment compared to not receiving treatment predicted significantly greater reductions in both probability bias (bStdYX = −.330, z = −3.645, p < .001) and cost bias (bStdYX = −.320, z = −3.495, p < .001). Second, treatment outcome was predicted by threat appraisals; higher residualized gain scores for social anxiety were predicted by both higher residualized gain scores for probability bias (bStdYX = .432, z = 4.433, p < .001) and cost bias (bStdYX = .285, z = 2.815, p = .005). Third, the mediators were significantly positively correlated (bStdYX = .625, z = 10.015, p < .001); and fourth, the effect of treatment on treatment outcome was statistically mediated by pre-to-post changes in both cost bias and probability bias. That is, the indirect a × b pathway was significant for the OPQ (bStdYX = −.143, z = −2.780, p = .005) and OCQ (bStdYX = −.091, z = −2.182, p = .029). Next 5000 bootstrap samples were generated to obtain the most accurate confidence intervals for indirect effects in mediation ( MacKinnon, Lockwood, & Williams, 2004). Neither the confidence interval for the OPQ [95% CI −.490, −.081] nor that for the OCQ [95% CI −.381, −.027] overlapped with zero, further supporting the finding of statistically significant mediation. Multiple mediators path model. BFNE=Brief Fear of Negative; OCQ=Outcome Cost ... Fig. 3. Multiple mediators path model. BFNE = Brief Fear of Negative; OCQ = Outcome Cost Questionnaire; OPQ = Outcome Probability Questionnaire. Parameter estimates are reported with standard errors in parentheses. * = Parameter estimate is significant at the .05 level; ** = Parameter estimate is significant at the .01 level; *** = Parameter estimate is significant at the .001 level. Figure options To test whether or not threat reappraisal causes anxiety reduction (Criterion 3), a cross-lagged panel design path model was employed to investigate the presumed causal interplay among social anxiety symptoms (BFNE) and cost and probability biases (OCQ, OPQ) at three time points: pretreatment, midtreatment, and posttreatment. The model incorporates autoregressive effects that control for temporal stability within threat appraisal and social anxiety scores across time, synchronous correlations between variables at each time point that account for covariances between variables not already explained by the influences of the variables from earlier time points, and cross-lagged direct effects. Thus any cross-lagged effects can be considered effects that add predictive power over and above that which can simply be obtained from within-construct stability over time and synchronous and other IV effects. The cross lags between social anxiety and judgmental biases at mid- and posttreatment are of primary importance to our study hypothesis, as they allow for evaluation of three potential scenarios: 1) whether earlier levels of judgmental biases predicted later changes in social anxiety, 2) whether earlier levels of social anxiety predicted later changes in judgmental biases, and 3) whether any relations were reciprocal. Social anxiety and judgmental biases at pretreatment are also important for this study, because their inclusion serves as a control for pretreatment symptom severity, thereby providing an indicator of change (Finkel, 1995 and Rieckmann et al., 2006). For example, social anxiety at midtreatment represents residualized change in social anxiety from pretreatment to midtreatment. The cross-lagged panel design model is presented in Fig. 4. Results indicated the fully cross-lagged model had acceptable fit according to all indices (Model χ2(13) = 22.053, p = .055; CFI = .974; TLI = .933; SRMR = .053), with the exception of the RMSEA (RMSEA = .096 [.000, .163], which is slightly above conventional standards for good fit ( MacCallum, Browne, & Sugawara, 1996). It should be noted, however, that the RMSEA tends to reject acceptable models when sample sizes are small. For this reason, some scholars argue against computing the RMSEA for models with low degrees of freedom ( Kenny, Kaniskan, & McCoach, 2014). Fig. 4 also shows the standardized path coefficients for the model. As predicted, examination of individual paths revealed significant autoregressive effects and intercorrelations between variables at each time point. The cross lag model revealed a significant effect of midtreatment OPQ on posttreatment BFNE (bStdYX = .350, z = 4.473, p < .001), specifically lower probability bias predicting lower social anxiety symptoms. However, the cross lag from midtreatment OCQ to posttreatment BFNE was not significant (bStdYX = −.059, z = −.719; p = .472); nor was the cross lag from midtreatment BFNE to posttreatment OPQ (bStdYX = .038, z = .419, p = .675) or from midtreatment BFNE to posttreatment OCQ (bStdYX = .049, z = .364, p = .716). Findings suggest that lower midtreatment levels of probability bias, but not cost bias, predicted greater reduction in social anxiety symptoms. That the inverse is not supported (i.e., that midtreatment BFNE does not predict change in OPQ) suggests the relation is not reciprocal and provides further evidence supporting probability bias as a specific cognitive reappraisal mediator of CBT for SAD. 1 Cross-lagged panel design path diagram relating the BFNE, OCQ, and OPQ. ... Fig. 4. Cross-lagged panel design path diagram relating the BFNE, OCQ, and OPQ. BFNE = Brief Fear of Negative; OCQ = Outcome Cost Questionnaire; OPQ = Outcome Probability Questionnaire. Standardized parameter estimates are reported with standard errors in parentheses. * = Parameter estimate is significant at the .05 level; ** = Parameter estimate is significant at the .01 level; *** = Parameter estimate is significant at the .001 level. Figure options To test the specificity of the relation between threat reappraisal and social anxiety reduction (Criterion 4), a second cross-lagged panel design path model was utilized to analyze the relation between social anxiety symptoms (BFNE), probability bias (OPQ), and working alliance (WAI), a plausible rival mediator, at three time points throughout treatment. Fit indices, again with the exception of the upper limit of the RMSEA confidence interval, indicated good model fit (Model χ2(14) = 18.705, p = .177; RMSEA = .066 [.000, .138]; CFI = .984; TLI = .963; SRMR = .037). Fig. 5 shows the standardized path coefficients for the model. Examination of individual paths revealed significant autoregressive effects between variables at each time point. However, working alliance was not significantly correlated with probability bias or with social anxiety symptoms at any time point. The cross lag model again revealed a significant effect of midtreatment OPQ on posttreatment BFNE (bStdYX = .352, z = 3.480, p = .001), whereas the cross lag from midtreatment WAI to posttreatment BFNE was not significant (bStdYX = −.050, z = −.657; p = .511). Findings suggest that earlier levels of probability bias, but not working alliance, predicted later change in social anxiety symptoms. These findings provide further support for threat reappraisal, specifically probability bias, as a mediator of CBT for SAD by demonstrating specificity of the threat reappraisal-social anxiety reduction relation. Cross-lagged panel design path diagram relating the BFNE, OPQ and WAI. ... Fig. 5. Cross-lagged panel design path diagram relating the BFNE, OPQ and WAI. BFNE = Brief Fear of Negative; OPQ = Outcome Probability Questionnaire; WAI = Working Alliance Inventory. Standardized parameter estimates are reported with standard errors in parentheses. * = Parameter estimate is significant at the .05 level; ** = Parameter estimate is significant at the .01 level; *** = Parameter estimate is significant at the .001 level.