بررسی امکان سنجی فعالسازی رفتاری تصادفی دادگاه برای واکنش آسیب شناسی سوگ

Abstract This study investigated the feasibility of using behavioral activation to treat enduring postbereavement mental health difficulties using a two-arm, multiple baseline design comparing an immediate start group to a delayed start group at baseline, 12-, 24-, and 36-weeks postrandomization. Participants received 12–14 sessions of behavioral activation within a 12-week intervention period starting immediately after the first assessment or after 12 weeks for the delayed start group. Prolonged grief, posttraumatic stress, and depression symptoms were assessed as outcomes. Compared with no treatment, behavioral activation was associated with large reductions in prolonged, complicated, or traumatic grief; posttraumatic stress disorder; and depression symptoms in the intent-to-treat analyses. Seventy percent of the completer sample at posttreatment and 75 percent at follow-up responded to treatment with 45 percent at posttreatment and 40 percent at follow-up being classified as evidencing high-end state functioning at 12-week follow-up.

.

Results Fidelity/Adherence As noted, adherence was monitored in weekly team meetings in which individual sessions were reviewed for adherence/fidelity and feedback provided before the next session. In addition, ratings of fidelity/adherence were completed using a version of the National Institute of Mental Health Collaborative Study Psychotherapy Rating Scale modified for BA at study end (CSPRS; Dimidjian, Herman-Dunn, Hubley, & Martell, 2009). Ten percent of the total number of sessions were randomly selected and rated by two raters, making sure every participant had at least one videotape rated. Raters were advanced Ph.D. students with experience with BA not involved with treating the participant (e.g., Hill, O’Grady, & Elkin, 1992). CSPRS training entailed reading and discussing the manual and then rating a series of practice sessions until near perfect reliability with the fidelity trainer was achieved. Average interrater reliability was good (ICC = .82) with the average overall rating being 5.22 (SD = .79; 5 = considerably) on a scale of 1 (not at all) to 7 (extensively). Intent-To-Treat Analysis We used a maximum likelihood estimation approach to fit two sets of two-level linear random coefficients models using SAS PROC MIXED for each of our outcomes (depression, PGD, and PTSD symptoms). These models examined (a) the effect of treatment at Time 2 by comparing the immediate start group at posttreatment assessment with the delayed start group just beginning treatment using a dummy-coded time variable, and (b) the effect of treatment over time by examining changes from pretreatment (T1 for immediate start, T2 for delayed start), posttreatment (T2 for immediate start, T3 for delayed start), to follow-up (T3 for immediate start, T4 for delayed start) across groups as they passed through the same stages of treatment. In both sets of analyses, individual observations over time were nested within treatment arms (immediate start vs. delayed start). In these analyses we used maximum likelihood estimation (MLE) to compensate for missing data given the increased efficiency of the estimation process, the need for fewer decisions by the researcher that might influence the outcomes, and the ability to replicate the results with the same data (see Allison, 2012, for review). Because standard error estimates in MLE is contingent on assumptions about the structure and stability of the covariance matrix, particular care is required in using MLE with small samples in order not to artificially constrain the standard error estimates and artificially inflate significance tests of the parameter estimates. To compensate for the potential instability of the estimates and the potential to underestimate the standard errors of the estimates due to small sample sizes, we took a conservative approach making no assumptions about the underlying covariance structure, using restricted MLE, and calculating parameter tests using Kenward-Roger corrections (Arnau et al., 2009, Giesbrecht and Burns, 1985 and Kenward and Roger, 1997). At this time, we are not aware of a consensus on how to calculate effect sizes in a multilevel random coefficients model (Roberts & Monaco, 2006). The effect size estimates reported below follow procedures similar to those reported in Oishi, Lun, and Sherman (2007), converting t values to d values using Formula 2.8 in Rosenthal, Rosnow, and Rubin (2000). We used adjusted dfs from the Kenward-Roger corrections in these estimates. Effect of Treatment Versus No Treatment The lagged start design of this study allowed for a control group comparison at Time 2. Using the multilevel molding approach described above, we examined difference between groups at Time 2 by dummy coding each assessment time with Time 1 as the reference, adding these terms to Level 1 of the model and looking at the interaction with the Level 2 treatment condition nesting variable. This approach allowed for estimation of the differential effect of completing treatment (immediate start group) versus no treatment (delayed start pretreatment) at Time 2 within the ITT sample by the Time 2dummy × Group coefficient. Results indicated large reductions in the immediate start group as compared with the delayed start group at Time 2 in symptoms of depression (γ11 = –10.00, p = .03, d = –1.01), PGD (γ11 = –11.05, p = .001, d = –1.74), and PTSD (γ11 = –13.09, p = .02, d = –1.08), supporting the feasibility of using BA within our target population. Effects of Treatment Over Time Results from the random coefficients models examining the effects of treatment over time are summarized in Table 2. Time was coded such that pretreatment was set at zero for both groups so that we could examine the change over time attributable to BA across both groups. The treatment arm group variable was dummy coded with the immediate start group set as the reference group to examine whether treatment after a delay differed from immediate start in the assessment that immediately preceded commencement of treatment. For both arms, self-reported depression symptom severity, PGD symptom ratings, and PTSD symptom ratings declined over the course of treatment (γ10 = –4.95, p < .02, ddepression = –1.19; γ10 = –7.53, p < .0001, dPGD = –1.83; and γ10 = –8.53, p < .002, dPTSD = –1.63, respectively). While rate of change during the treatment phase was not contingent on assignment to treatment arm for depression ratings (γ11 = 3.71, p > .10), PGD ratings (γ11 = 1.51, p > .10), or PTSD ratings (γ11 = 2.75, p > .10), the delayed treatment group did evidence effects of maturation in marginally significant reductions in depression (γ01 = –7.81, p = .07) and PTSD symptom ratings (γ01 = –8.26, p = .10) at pretreatment compared with the immediate start group. To explore this, we examined the pretreatment scores (T2) in the delayed treatment group and found one person who no longer met criteria for PGD based on the relevant items from the ICG-R. This change was due to the person reporting three problems interfering with functioning rather than the requisite five. This person also did not meet clinical cutoffs on the other two outcome scales at treatment initiation (pre-treatment PG score = 27, DASS depression score = 5, PCL-S score = 26). Table 2. Results of Multilevel Random Coefficients Models for Self-Reported Depression, PGD, and PTSD Symptoms for Pretreatment, Posttreatment, and Follow-up Collapsed Across Groups Effect Unstandardized coefficient SE df t p d Depression symptoms Intercept, b0 Intercept, γ00 14.29 3.08 21 4.64 .001 2.03 Group, γ01 7.81 4.10 21.1 1.91 .07 .83 Change over time, b1 Time, γ10 –4.95 1.86 19.8 –2.65 .02 –1.19 Group × Time, γ11 –3.71 2.50 20 –1.48 .15 –.66 PGD symptoms Intercept, b0 Intercept, γ00 35.56 2.18 20.5 16.34 < .0001 7.22 Group, γ01 4.31 2.90 20.5 1.49 .15 .66 Change over time, b1 Time, γ10 –7.53 1.32 39.1 –5.71 < .0001 –1.83 Group × Time, γ11 –1.51 1.77 39.3 –.85 .40 –.27 PTSD symptoms Intercept, b0 Intercept, γ00 40.52 3.63 21 11.17 < .0001 4.87 Group, γ01 8.26 4.83 21 1.71 .10 .75 Change over time, b1 Time, γ10 –8.53 2.35 19.8 –3.63 .002 –1.63 Group × Time, γ11 –2.75 3.16 20 –.87 .39 –.39 Note. Group variable coded as delayed = 0, immediate = 1. PGD = prolonged, complicated, or traumatic grief; PTSD = posttraumatic stress disorder. Table options Clinical Significance and Acceptability in Completer Sample Five participants dropped out of the study after randomization; two from the immediate start group and three from the delayed start group. Of these, three dropped out before the first treatment session, and two during the first three sessions of treatment. Completers did not differ from noncompleters in terms of age, t(22) = .80, p > .05 (Mcompleters = 50.45, Mnoncompleters = 44.25); self-reported PTSD, t(23) = –.65, p > .05 (Mcompleters = 51.05, Mnoncompleters = 55.29); or self-reported PGD, t(23) = –1.23, p > .05 (Mcompleters = 40.77, Mnoncompleters = 44.52). There were significant differences in days since death, t(22) = –.65, p < .05 (Mcompleters = 770.00, Mnoncompleters = 1977.50), though given the small sample size and one noncompleter having a score more than three standard deviations above the mean for the sample (3,630 days) may account for this finding. There were no significant differences between dropouts and the rest of the sample on any outcome at Time 1 (t values between .42 and .71). However, across outcome measures noncompleters reported somewhat higher levels of symptoms. Both groups were well above established clinical cutoffs for PTSD (clinical cutoff for PCL-S = 44; Blanchard, Jones-Alexander, Buckley, & Forneris, 1996). However, the completer groups fell within the severe range based on DASS depression subscale norms, while the noncompleter groups fell into the extremely severe range on average ( Lovibond & Lovibond, 1995). There were no differences in incidence of preloss mental health diagnoses, LR(1, N = 25) = 1.35, p > .05, nor use of psychotropic mediations, LR(1, N = 25) = 1.66, p > .05. Across conditions, pretreatment scores on the modified ICG-R used in our study went from M = 39.27 to M = 25.12 at posttreatment. Mean depression and PTSD symptom ratings decreased from 19.83 on the DASS-42 depression subscale (moderate range; Crawford & Henry, 2003) and 47.10 on the PCL-S (clinical cutoff of 44; Blanchard et al., 1996) to 6.70 (normal range) for depression and 28.42 for PTSD. Those participants evidencing a 20% reduction in symptoms in all outcome measures from pretreatment to posttreatment and also at follow-up were classified as a treatment responder at that time point ( Borkovec et al., 2002, Ladouceur et al., 2000 and Roemer and Orsillo, 2007). In this study, 70% of the completer sample evidenced a 20% reduction in self-reported PGD, depression, and PTSD symptoms at posttreatment and 75% evidenced a 20% reduction at the 12-week follow-up. Given no published norms for the modified ICG-R used in this study, analysis of high-end state functioning was based on published norms for MDD and PTSD symptoms in nonclinical populations using DASS ( Crawford & Henry, 2003) and PCL-S ( Walker, Newman, Dobie, Ciechanowski, & Katon, 2002) scores. These analyses indicated that 45% of the completer sample at posttreatment and 40% of this sample at follow-up met criteria for high-end state functioning (DASS depression < 9, PCL-S < 26). Using the conservative Jacobson-Truax method (Atkins et al., 2005 and Jacobson and Truax, 1991) to examine clinically significant change in individual outcome measures, we found that despite large gross changes in overall symptom levels there were differential levels of response to treatment based on outcome (Table 3). This method entails a two-step criterion for clinically significant change. First, a cutoff between the patient/dysfunctional and nonpatient/functional populations is established, defined as the point two standard deviations below the range of the pretherapy mean for the outcome measure examined (Cutoff A; 4.98 for DASS depression scores, 26.17 for PCL-S scores, and 29.19 for ICG-R scores in this study). Next, a Reliable Change Index (RCI) for each participant is calculated to ensure that changes are not an artifact of measurement error (RCI = (MEANposttreatment – MEANpretreatment)/STANDARD ERRORdifference). The lower bounds for indicated change from the RCIis 1.96. Based on the two-step criterion, individuals will be classified as recovered (passed both Cutoff A and RCI criteria), improved (pass RCI criterion but not Cutoff A), unchanged (passed neither criteria), or deteriorated (passed RCI criterion but symptom scores increased) for each interval. Specifically, 95% could be classified as recovered or improved in terms of PGD symptoms at follow-up, 75% in terms of MDD symptoms, but a relatively more modest 45% met these criteria for PTSD symptoms. Table 3. Classifications at Posttreatment and at 12-Week Follow-up Based on Jacobson-Truax Cutoff A and Reliable Change Index Treatment response Recovered Improved Unchanged Deteriorated PGD Posttreatment 45% 45% 10% 0% Follow-up 60% 35% 5% 0% Depression Posttreatment 65% 0% 35% 0% Follow-up 75% 0% 25% 0% PTSD Posttreatment 0% 55% 45% 0% Follow-up 20% 25% 55% 0% Note. PGD = prolonged, complicated, or traumatic grief; PTSD = posttraumatic stress disorder. Table options The intervention was well tolerated with dropout (20%) at similar levels as seen in other BA treatment studies (Cuijpers et al., 2007). Posttreatment CEQ ratings indicated that the treatment was credible and participants were satisfied with the treatment they received. Using a cutoff of seven or more on CEQ items as an indicator of high satisfaction with treatment, we found that 95% rated the intervention as logical, 85% as successful, and 100% would confidently recommend this intervention to another. Ratings of subject sense of improvement in symptoms immediately at posttreatment included one person reporting a 30% improvement, two reporting a 40% improvement, one a 50% improvement, five a 70% improvement, six 80%, and seven a 90% or more improvement.