A preliminary multivariate analysis of variance (MANOVA) for PSS and BDI pretreatment scores indicated a significant effect of diagnosis, F(2,91)=12.78, p=.000, η2=0.22. Follow-up analyses demonstrated that the comorbid group had significantly higher scores than the pure PTSD group on the PSS, F(1, 92)=19.13, p=.000, η2=0.17, and on the BDI, F(1, 92)=20.18, p=.000, η2=0.18. Chi-square analyses indicated the treatment groups were comparable in terms of comorbidity, and that drop-out was proportional in terms of treatment group and comorbidity (in the CPT group, 10 PTSD and 7 PTSD+MDD participants dropped out of therapy, and 9 and 5 in the PE group, respectively). Reported degrees of freedom vary due to missing data.
A 2 (group: CPT, PE) × 2 (diagnosis: PTSD only, PTSD+MDD) multivariate analysis of covariance (MANCOVA) was conducted for posttreatment guilt scores and controlled the effects of pretreatment guilt levels. Descriptive statistics are reported in . There was a main effect of group, F(4, 75)=5.33, p=.001, η2=0.22, but not for diagnosis, F(4, 75)=0.58, p=.68, η2=0.03, nor group by diagnosis interaction, F(4, 75)=0.37, p=.83, η2=0.02.
Means, standard deviations and effect sizes at pre-, post- and 9-month follow-up assessments for cognitive-processing therapy (CPT) and prolonged exposure (PE)
Planned comparisons were conducted by dividing the sample into four groups based upon their diagnostic status at pretreatment, those with pure PTSD who had received CPT (CPT: PTSD only), those who had comorbid MDD and received CPT (CPT: PTSD+MDD), and repeated this stratification for the PE condition (i.e., PE: PTSD only, PE: PTSD+MDD). Pretreatment guilt levels were controlled. For the pure PTSD group, CPT participants reported significantly lower hindsight bias scores, F(1, 49)=8.68, p=.005, η2=0.15, and lack of justification scores, F(1, 48)=6.46, p=.014, η2=0.12, than PE participants. The two groups were comparable in terms of global guilt and wrongdoing scores, F(1, 52)=0.91, p=.35, η2=0.02, and F(1, 50)=2.53, p=.12, η2=0.05, respectively. A similar pattern of results was evident for the comorbid group: hindsight bias, F(1, 39)=4.22, p=.047, η2=0.10; lack of justification scores, F(1, 38)=9.14, p=.004, η2=0.19; global guilt, F(1, 40)=2.01, p=.17, η2=0.05; wrongdoing scores, F(1, 37)=1.01, p=.32, η2=0.03. These results were therefore contrary to the hypothesis that the larger effects of CPT over PE in treating trauma-related guilt were driven by a subset of depressed participants. Controlling for pretreatment PSS and BDI scores did not alter these findings.
A 2 (group: CPT, PE) × 2 (diagnosis: PTSD only, PTSD+MDD) MANCOVA was conducted to examine follow-up guilt scores (with pretreatment guilt levels being controlled). Main effects of group and diagnosis were nonsignificant, F(4, 52)=1.61, p=.19, η2=0.11, and F(4, 52)=0.84, p=.51, η2=0.06, as was the group by diagnosis interaction, F(4, 52)=0.25, p=.91, η2=0.02. Planned comparisons did not reveal any significant findings. Accordingly, participants with pure PTSD in both groups had comparable levels of guilt: global guilt, F(1, 40)=2.34, p=.13, η2=0.06; hindsight bias, F(1, 37)=1.37, p=.07, η2=0.08; lack of justification scores, F(1, 35)=0.34, p=.57, η2=0.01; wrongdoing scores, F(1, 38)=0.38, p=.54, η2=0.01. Findings for the comorbid group were essentially the same: global guilt, F(1, 25)=0.51, p=.48, η2=0.02; hindsight bias, F(1, 25)=1.87, p=.18, η2=0.07; lack of justification scores, F(1, 26)=3.92, p=.058, η2=0.13; wrongdoing scores, F(1, 25)=0.10, p=.75, η2=0.00.
3.1. Clinical significance
We then examined the proportion of participants who made reliable and clinically significant changes in trauma-related guilt following treatment as outlined by Jacobson and Truax (1991)
, which had not been done in the original report. Jacobson and Truax (1991)
define a reliable change by a change of more than 1.96 S.E.M.s between pretreatment and posttreatment (or follow-up). Individuals were considered to be in the clinical range at pretreatment if they had a mean guilt score of 1.75 or more (and were therefore included in the analysis), and a mean score of 1.0 or less was considered to reflect minimal levels of guilt at posttreatment (and follow-up) (E.S. Kubany, personal communication, June 28, 2004). Due to small cell sizes, Fisher’s Exact Test was used. Effect sizes are reported as phi coefficients (ϕ), where .10 is considered small, .30 medium, and .50 large (Cohen, 1988
). As indicated in , greater proportions of CPT than PE participants made clinically significant changes on guilt measures, independent of comorbidity status. Examination of effect sizes suggested that more significant findings favouring CPT would have been observed with increased sample size.
Proportion of clients who made clinically significant changes at posttreatment and 9-month follow-up