**2. Method**

### **2.1. Search strategy**

Studies published from 1980 (with the introduction of PTSD in the DSM-III) to the end of December 2012 were identified by searching the electronic databases MedLine, PsycINFO, PILOTS, and ProQuest Dissertations and theses databases [19] to identify relevant studies evaluating the impact of CBT and Prazosin on PTNMs in an adult population. Each search was separately run using the following command in all fields: (treatment OR therapy OR Prazosin OR cognitive OR behav\* OR CBT) AND (nightmare\* OR dream\*) with the asterisk specifying the plural and grammatical variations. We also scanned reference lists of relevant metaanalyses, systematic reviews, and reference lists of each study included in this meta-analysis.

#### **2.2. Eligibility criteria**

To be included in the meta-analysis, studies must have met the following criteria: (a) they reported quantitative treatment outcomes for NMs (such as frequencies or questionnaires' scores) occurring after a traumatic event; (b) at least one group received CBT or Prazosin treatment for NM; (c) the study either had a group protocol or was a case study of at least three participants; (d) the participants were 18 years old and above; (e) all the participants presented PTSD symptoms (evaluated by validated questionnaires or by a clinical and structured interview) and PTNM complaints; (f) the study was published between 1980 and 2012; and (g) the study was published in English or in French. Studies were excluded if they were (a) abstracts of posters and (b) did not report original results.

#### **2.3. Data extraction**

Information was extracted from each eligible record using a data extraction sheet adapted from previous meta-analyses [11, 20]. Data extracted were : (i) study identification (e.g., authors and publication date); (ii) study description (e.g., protocol sampling, assignment of participants, and research reports following CONSORT statement for randomized studies [21]); (iii) sample characteristics (i.e., sample size, gender, mean age and standard deviation, PTSD and acute stress disorder (ASD) diagnoses, records of PTSD treatments, trauma type, trauma date of occurrence, nightmare definition, nightmare frequency, nightmare intensity, nightmare content, explicit report of sleep difficulties other than nightmares, reports of comorbidity, and attrition rate); and (iv) the type of intervention for NMs at large was also reported (i.e., IRT, IRET, ERRT, LDT, relaxation, exposition, desensitization, EMDR, Prazosin, and any other CBTs for NMs). CBTs for NMs were defined as incorporating cognitive (new dream rehearsal, restructuration, etc.) and/or behavioral strategies (progressive muscle relaxation, gradual exposure, etc.), contrasting with other approaches, such as dream interpretation. We recorded the number of therapy sessions, duration, the therapy modality (i.e., individual psychotherapy, in groups, at distance with or without any contact), the presence of a therapy protocol, the therapist's training (psychologist, graduates in psychology, other medical professionals, other), and whether the CBT treatment also targeted PTSD and insomnia problems in addition to NMs. For Prazosin studies, trade names of Prazosin (Minipress, Vasoflex, Pressin, and Hypovase) were considered in the study screening and we extracted the average prescription (in mg/day), the timing of the prescription, the presence of a protocol for the prescription, and the presence of a washout period at follow-up. (v) In addition, the type of NMs, PTSD and insomnia outcome and measures were coded as either (a) clinic interview, (b) daily selfmonitory, (c) EEG for NMs and insomnia, (d) validated questionnaires, (e) subscales from questionnaires, (f) isolated items from questionnaires or interviews, or (g) survey. And, finally, (vi) information was gathered to calculate the following three types of effect sizes: group comparison on post-treatment data (between-group analysis), pre- vs. post-treatment data for

groups receiving CBT or Prazosin (within group-group analysis), and pre- vs. post-treatment data for control groups.

#### **2.4. Inter-rater agreement**

the plural and grammatical variations. We also scanned reference lists of relevant metaanalyses, systematic reviews, and reference lists of each study included in this meta-analysis.

To be included in the meta-analysis, studies must have met the following criteria: (a) they reported quantitative treatment outcomes for NMs (such as frequencies or questionnaires' scores) occurring after a traumatic event; (b) at least one group received CBT or Prazosin treatment for NM; (c) the study either had a group protocol or was a case study of at least three participants; (d) the participants were 18 years old and above; (e) all the participants presented PTSD symptoms (evaluated by validated questionnaires or by a clinical and structured interview) and PTNM complaints; (f) the study was published between 1980 and 2012; and (g) the study was published in English or in French. Studies were excluded if they were (a)

Information was extracted from each eligible record using a data extraction sheet adapted from previous meta-analyses [11, 20]. Data extracted were : (i) study identification (e.g., authors and publication date); (ii) study description (e.g., protocol sampling, assignment of participants, and research reports following CONSORT statement for randomized studies [21]); (iii) sample characteristics (i.e., sample size, gender, mean age and standard deviation, PTSD and acute stress disorder (ASD) diagnoses, records of PTSD treatments, trauma type, trauma date of occurrence, nightmare definition, nightmare frequency, nightmare intensity, nightmare content, explicit report of sleep difficulties other than nightmares, reports of comorbidity, and attrition rate); and (iv) the type of intervention for NMs at large was also reported (i.e., IRT, IRET, ERRT, LDT, relaxation, exposition, desensitization, EMDR, Prazosin, and any other CBTs for NMs). CBTs for NMs were defined as incorporating cognitive (new dream rehearsal, restructuration, etc.) and/or behavioral strategies (progressive muscle relaxation, gradual exposure, etc.), contrasting with other approaches, such as dream interpretation. We recorded the number of therapy sessions, duration, the therapy modality (i.e., individual psychotherapy, in groups, at distance with or without any contact), the presence of a therapy protocol, the therapist's training (psychologist, graduates in psychology, other medical professionals, other), and whether the CBT treatment also targeted PTSD and insomnia problems in addition to NMs. For Prazosin studies, trade names of Prazosin (Minipress, Vasoflex, Pressin, and Hypovase) were considered in the study screening and we extracted the average prescription (in mg/day), the timing of the prescription, the presence of a protocol for the prescription, and the presence of a washout period at follow-up. (v) In addition, the type of NMs, PTSD and insomnia outcome and measures were coded as either (a) clinic interview, (b) daily selfmonitory, (c) EEG for NMs and insomnia, (d) validated questionnaires, (e) subscales from questionnaires, (f) isolated items from questionnaires or interviews, or (g) survey. And, finally, (vi) information was gathered to calculate the following three types of effect sizes: group comparison on post-treatment data (between-group analysis), pre- vs. post-treatment data for

abstracts of posters and (b) did not report original results.

208 A Multidimensional Approach to Post-Traumatic Stress Disorder - from Theory to Practice

**2.2. Eligibility criteria**

**2.3. Data extraction**

In order to evaluate inter-rater reliability and to control for data extraction bias, each study was systematically checked by pairs of independent reviewers. Their results were compared and any disagreements were resolved by discussion. Before beginning the coding process, the rating instrument was tested with the four raters going through the coding form and checking one article testing CBT and one trial studying Prazosin together. Judges were students enrolled in a PhD program in psychology. The inter-rater reliability was good and varied from 58.3% to 100%. The strongest agreements were related to sampling method and treatment type while the lowest referred to attrition and female percentages, which may reveal inconsistencies in reporting. The final sample was composed of 26 studies published in English: 8 studies on Prazosin [22–29] and 18 studies on CBT for NM [6, 30–46] (**Figure 1**).

**Figure 1.** Search strategy flowchart.

#### **2.5. Data analyses**

The meta-analyses were performed using the random-effects model [47]. The analyses were conducted with *Comprehensive Meta-Analysis* (*Version 2*) [48]. Effect sizes (ES) were computed with Hedges' *g* between groups using mean and standard deviation (*SD*) [47] except for one study where they were obtained from *t* values [37]. When available, quantitative analyses were performed on the difference between the experimental and control groups at posttreatment (between-group); otherwise analyses were computed on pre- and post-data for the same group (within-group). As in the last case, the correlation coefficient *r* between pre- and post-test for each data is needed, and as this information was missing, *r* was set to 0 in order to be conservative and to give to these studies the smallest and the same weight, so there will be no bias in weighting [47]. However, we executed sensitivity analysis using a range of plausible correlations (0; 0.5 and 0.95) and the results did not significantly differ between each correlation. The direction of the ES was set as positive when NMs, PTSD symptoms, and sleep variables were improved at post-treatment or for the experimental compared to the control group at post-treatment. ES were not computed at follow-up due to disparities in reporting these data for CBT studies, and only one Prazosin study provided them.

For the NMs variable, effect sizes were calculated based on means and *SD* of weekly NM frequency for CBT studies; and on the CAPS-B2 final scores (item B2 on the *Clinician Administered PTSD Scale*) for Prazosin studies. The CAPS-B2 item sums up the frequency and the intensity (from 0 to 4) of recurrent distressing dreams related to the traumatic event (refer to Blake et al. [49] for a scale description). When these NMs were reported by month or by night, we computed their mean by the week. Effect sizes were also computed for PTSD symptoms. For CBTs, we considered any questionnaire evaluating the construct of PTSD symptoms intensity: *impact of event scale* (IES-R [50]), PTSD *Checklist—Civilian and Military Versions* (PCL [51]), *post-traumatic symptom scale* (PSS [52]), and *post-traumatic diagnostic scale* (PDS [53]). For Prazosin, we used the scores of the PCL and the CAPS. Finally, ES were calculated for sleep difficulties using (1) the CAPS-D1 (CAPS, item D1) final score (sum of the frequency and intensity) for Prazosin and (2) scores on the *Pittsburgh Sleep Quality Index* (PSQI) [54] for CBTs. Computations were also possible for the insomnia construct, as data Pertained only to CBT studies, using scores on the *Sleep Impairment Index* (SII [55]).

We used the data, if available, on an intent-to-treat basis (patients who were enrolled and randomly allocated to treatment), otherwise variables were selected on a completer basis. Also, as a few study designs did not implement immediate post-evaluation and rather evaluated the interventions at follow-up, we decided to include them in the analysis if follow-ups were less than three months after the end of the treatment. Each effect size (and the combined effect size) was calculated with 95% confidence intervals.

We tested for heterogeneity with the *Q* test and the proportion of true variance was assessed with the *I*<sup>2</sup> index for each computed ES. *Q* tests were also used for contrast analyses. Also, due to problems of power in this meta-analysis (less than 20 studies) (*n* = 8 for Prazosin; *n* = 18 for CBT studies), we computed the 95% confidence intervals (95% CI) for the *I*<sup>2</sup> index [56, 57]. Heterogeneity was detected when *Q* test was significant (*p* < 0.05) or *I*<sup>2</sup> index was higher than 50% [56]. In this case, sensitivity analyses and ES comparison across subgroups (subgroup analyses) were performed using the random effect model. Finally, to test for possible publication biases, funnel plots evaluating the association between the Hedges' *g* and the standard error were visually assessed [58]. In addition, Orwin fail-safe N tests [59] were computed to estimate the number of studies with no effect necessary to reduce the combined ES to a clinically nonsignificant value (0.2).
