Summations Our network meta-analysis did not identify highly divergent levels of efficacy and acceptability among psychotherapies. Meta-cognitive therapy and cognitive processing therapy were nonetheless the most efficacious of the included psychotherapies. Narrative exposure therapy and written exposure therapy had relatively high rates of treatment completion when considered alongside other psychotherapies. Limitations We included psychotherapy approaches which have been subject to few trials and which await replication efforts from independent investigators. Our review was limited to individual face-to-face psychotherapies. 1 INTRODUCTION

Post-traumatic stress disorder (PTSD) is characterised by exposure to trauma and symptom clusters including re-experiencing of the traumatic event, avoidance of stimuli associated with the trauma, negative alterations in cognition and mood, and hyperarousal.1 It is common (8.3% lifetime prevalence2) and associated with multiple negative outcomes, including functional impairment,3 suicidality,4 co-occurring psychological disorders5, 6 and physical morbidity.7, 8

There is an apparent consensus within the PTSD field that trauma-focused psychotherapies are, for the most part, comparable so far as efficacy and adherence are concerned.9-11 In some respects, this is surprising. Even though all trauma-focused psychotherapies are thought to involve a meaningful processing of the trauma memory in some respect, the proposed underlying mechanisms of trauma-focussed approaches, and indeed, the specific approaches of each respective intervention nonetheless vary. Comparable efficacy and adherence of trauma-focused psychotherapies have been supported by meta-analytic reviews that consistently suggest that cognitive behaviour therapy (CBT), exposure therapies and eye movement desensitisation and reprocessing (EMDR) are similarly effective in treating PTSD.9, 11-14 A meta-analysis by Ourgin14 found no difference between cognitive therapy (CT) and exposure therapy; however, findings were analysed from only five studies. A Cochrane review15 reported no statistically significant differences between trauma focussed-cognitive behaviour therapy (TF-CBT), EMDR and Stress Management post-treatment. Further, the authors reported that TF-CBT, EMDR and CBT were more effective than other therapies (ie non-directive, supportive, person-centred counselling, hypnotherapy and psychodynamic therapy). A comprehensive meta-analysis by Watts et al.11 reported effect sizes indicating superior efficacy for cognitive processing therapy (CPT; g = 1.69), exposure and CT (g = 1.52) and prolonged exposure (PE; g = 1.38).

The above reviews have informed the development of treatment guidelines and most guidelines recommend TF-CBTs. These are considered to include CT, CPT and PE therapy, as well as EMDR.16-19

However, the evidence base has been built upon studies which have rarely included more than two treatment arms and meta-analytic methodologies have not, until recently, been able to incorporate indirect evidence across studies which compare different combinations of approaches. For instance, the above-mentioned reviews of Bisson et al.15 and Watts et al.11 relied on evidence derived only from a series of separate head-to-head comparisons, limiting their findings. Therefore, the assumption of equality of outcomes across trauma-focused psychotherapies may have been prematurely conferred. In contrast, a full account of the evidence base would additionally consider indirect evidence, whereby the superiority of a given intervention can be ascertained even in instances where direct head-to-head comparison studies have not been conducted.

Network meta-analysis (NMA) holds particular promise when considering the relative efficacy of a broad range of psychotherapies. The approach provides a statistical methodology for evidence synthesis that can integrate both direct and indirect evidence from multiple treatment comparisons to estimate the inter-relationships across treatments.20 The NMA approach may be particularly useful when synthesising a clinical literature characterised by a broad array of different therapeutic modalities and ‘brands’ which have not always been directly compared—such as PTSD.

To date, there have been three NMAs exploring psychotherapeutic efficacies for adults diagnosed with PTSD. However, a significant limitation of each of these has been the lumping of multiple separate psychotherapies into TF-CBT or ‘psychotherapy’ clusters, thus calling for a NMA whereby the relative efficacy and adherence of individual psychotherapies can be ascertained. For instance, Gerger et al.10 synthesised data from 66 randomised controlled trials (RCTs) and reported no superior interventions, with similar efficacy reported for CBT, CT, EMDR and exposure therapy. However, this NMA did not define a distinction between CBT, exposure therapies, or CT and included these all within a TF-CBT framework for the final NMA model, reducing the granularity and clinical utility of these results. Finally, publications were restricted to 1980 until 2010, such that recent RCTs were not included.

The second NMA by Merz, Schwarzer and Gerger21 examined the efficacy and acceptability of psychotherapies, pharmacotherapies or combined treatment approaches, at post-treatment and long-term follow-up. In this context, acceptability referred to the proportion of participants completing treatment rather than dropping out due to adverse effects of the intervention. Their results indicated a superiority of psychotherapies over pharmacotherapy at long-term follow-up, and no differences in acceptability across comparisons. This study utilised data from only 12 RCTs, with long-term follow-up not reported for the full sample (six studies only). Most importantly, however, the findings do not allow a comparison of the relative efficacy of each individual psychotherapy, as each approach was analysed under the broad umbrella of ‘psychotherapy’. Thus, CBT, PE, EMDR and Seeking Safety (a CBT- and psychodynamic-derived group therapy for PTSD and co-morbid substance use that proscribes exploration of trauma memories22) were analysed as a unitary intervention category, precluding an assessment of psychotherapeutic (non)similarity.

A third NMA was recently reported by Mavranezouli and colleagues.23 These authors delineated EMDR from TF-CBT, but did not distinguish between individual TF-CBTs, such as CPT and exposure-based interventions. Also, Mavranezouli et al.23 did not consider acceptability in their analyses. Acceptability would appear to be a vital outcome to understand with regard to the non-distinctiveness of TF-CBTs, given that the common concern that exposure-based approaches may be associated with high rates of treatment dropout and discontinuation.

The large number of meta-analytic reviews has missed an opportunity to truly integrate the sizable body of PTSD psychotherapeutic literature and to examine whether some therapies are indeed more efficacious and acceptable than others. The NMA approach holds great promise for identifying the most efficacious and acceptable interventions for PTSD; however, the application of NMA has likewise, until now, overlooked the opportunity to identify whether some TF-CBTs are more efficacious and acceptable than others. Therefore, in this study, the aim was to conduct a comprehensive systematic review and NMA to inform clinical practice by comparing psychotherapies to evaluate their efficacy on PTSD symptom reduction, as well as their acceptability defined as all-cause discontinuation.24

2 METHOD

The review was registered with the International prospective register of systematic reviews (PROSPERO CRD42019119814), and we report the results consistent with the PRISMA extensions statement for NMA.25

2.1 Search strategy and selection criteria

A comprehensive and systematic literature search was conducted to determine studies for inclusion for the NMA. The databases of EMBASE, PsychINFO, PTSDPubs and PubMed were searched from their inception to 21 January 2020. Search terms included MeSH terms as well as search terms such as “post-traumatic stress disorder” AND “psychotherapy” AND “ramdomi?ed controlled trial”. See Supplementary Material S1 for specific search term strategies. The principal investigator (BJ) and second author (AL) independently screened all titles and abstracts of studies identified by the electronic search. The full texts of any study deemed potentially eligible (n = 220) were then also independently screened by investigators BJ and AL, and disagreement was resolved by discussion. See Supplementary Material S2 for the excluded trials.

We included RCTs comparing bona-fide psychotherapies to control (ie waitlist), or another psychotherapy, for the treatment of adults (≥18 years old and of both sexes), with a primary diagnosis of PTSD according to the standard operationalised diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders (DSM-III-R26; DSM-IV-TR27; DSM-IV28; and DSM-529) and the International Classification of Diseases and related health problems (ICD-930; ICD-1031). Studies derived from DSM-III1 criteria were not included due to significant changes to the PTSD diagnostic criteria from DSM-III1 to DSM-III-R,26 as well as publication date, which may have violated the transitivity assumption. Two studies were excluded for this, and other reasons (see Supplementary Material S2 for further details). Papers were required to be in English and published in a peer-reviewed journal. In contrast to the stated PROSPERO registration (CRD42019119814), we did not require adherence of primary studies to the CONSORT statement as we were not always able to reliably determine whether this was the case. So far as the description of therapies is concerned, at least two of the following had to be fulfilled for study inclusion: (a) a citation to an established school or approach to psychotherapy, (b) a description of the therapy that contained a reference to a psychological process (eg operant conditioning), (c) a reference to a treatment manual that was used to guide the delivery of the treatment and/or (d) the identification of active ingredients. See Supplementary Material S1 for definition and description of bona-fide psychotherapies. We considered only individual, face-to-face psychotherapies (eg excluding group or conjoint interventions) to improve methodological rigour and to help inform practitioners working with individuals in routine treatment settings, where trauma-focused therapies involving the processing of idiosyncratic trauma memories are typically administered in an individual therapy context. Additionally, we only included studies which reported a direct comparison between at least two individual psychotherapy modalities, or between a psychotherapy and control condition, and we thus excluded studies of sequential treatments if direct comparison data between individual interventions could not be extracted. A variety of active and waitlist conditions have been utilised in the literature. The control conditions were combined into four separate nodes (waitlist, active supportive therapy, treatment-as-usual and psychoeducation). Supplementary Material S1 highlights the key features of each unique control condition which informed the categorisation process.

2.2 Outcomes

The primary outcome was efficacy, represented by PTSD symptomatic change from baseline to post-treatment on a validated scale for PTSD. Long-term follow-up was outside the scope of the project. When PTSD symptoms were measured with more than one validated rating scale, a predefined hierarchy based on psychometric properties and consistency of use across included trials was utilised (see Supplementary Material S1). More frequently used scales and self-rated PTSD symptoms were preferred. Self-rated scales were favoured over interview-based measures given that proportionately few studies administered interview-based measures at post-treatment and because self-report measures do not have the addition of interviewer-based variance beyond respondent-based variance alone. Results from intention-to-treat (ITT) analyses were preferred over results from completer analyses. Between-group post-treatment standardised mean differences (SMD) with 95% confidence intervals (95% CIs) were calculated, with an effect of 0.24 to imply clinical importance.32

A secondary outcome was acceptability. Consistent with previous network meta-analytic reviews,21, 24, 33 acceptability refers to treatment discontinuation calculated by the proportion of patients who withdrew for any reason. Favourable acceptability reflects the completion of therapy or absence of discontinuation, and encompasses the notions of treatment adherence and non-attrition. Odds ratios (OR) with 95% CIs were used as a measure of the association between the psychotherapeutic approach and acceptability.

2.3 Data extraction

Data were extracted from the included studies by BJ using a structured form. Sample sizes, baseline and end of treatment means, and standard deviations (SDs) were extracted for effect size calculation for each treatment group. Where CIs were reported, conversions to SDs were calculated according to formulas provided by Higgins and Green.34 Authors of 22 studies were contacted due to insufficient data reported in the primary paper, with authors of four studies providing the relevant information.35-38 In addition to the data required for effect size calculation, other characteristics of trials were also extracted to identify potential effect modifiers, these being index trauma type (eg interpersonal violence), year of publication and diagnostic criteria.

2.4 Risk of Bias

The Cochrane Risk of Bias assessment tool was used to assess the quality of included studies.39 The assessment of the risk of bias included a brief training period whereby BJ and a co-rater reached consensus with a random sample of selected studies (n = 10), with the remainder being assessed by BJ. In addition, the Confidence in Network Meta-Analysis (CINeMA) platform was used to evaluate the quality of studies across the network.40, 41

2.5 Review of the network geometry

Published RCTs including patients with PTSD were analysed. In the network, each psychotherapy is indicated by a node, and comparisons between psychotherapies are shown by the links between the nodes.

2.6 Data analysis

For the primary analyses, SMD were estimated for continuous efficacy outcomes and ORs for dichotomous attrition outcomes using pairwise comparisons and NMA. The study effect sizes were then synthesised using a random-effects NMA model. Frequentist network meta-analyses were conducted using CINeMA, which integrates the R netmeta package.40 In the NMA, the heterogeneity variance parameter was assumed to be the same for all treatment comparisons. Heterogeneity was considered low with a value of τ2 = 0.04, moderate as τ2 = 0.09 and high heterogeneity as τ2 = 0.16.42

Consistency, that is, the agreement between direct and indirect evidence was statistically evaluated using a global design-by-treatment interaction model,43 and locally, by separating direct evidence from indirect evidence.44 The transitivity assumption was further evaluated by comparing the distribution of methodological variables that could act as effect modifiers across treatment comparisons in a pairwise meta-regression. A Knapp-Hartung method of meta-regression was conducted with separate estimates of τ2 for each subgroup.42 Estimates of τ2 were calculated using the maximum likelihood method.42 Waitlist was utilised as the standard comparator due to the condition being the most commonly compared control (n treatment arms = 57). To rank the treatment, each therapy was plotted for each NMA estimate variable compared to waitlist.44 To determine whether the network was susceptible to small-study bias, a funnel plot comparing studies with waitlist was produced, with asymmetry assed with the Egger's regression test.45

3 RESULTS 3.1 Results of the search

Results of the systematic search identified 5649 citations. Independent screening of titles and abstracts yielded high agreement between raters (Cohen's Kappa = .94; 99.32% agreement), resulting in 220 eligible articles retrieved in full-text (see Figure 1 for PRISMA diagram). Screening full-text articles yielded high inter-rater reliability (Cohen's Kappa = .95; 97.73% agreement). Overall, 82 RCTs were included in the analysis, comprising 5775 participants (see Supplementary Material S1 for full reference list). The included studies were conducted between 1991 and 2020, comparing 17 psychotherapies and four control conditions.

PRISMA diagram of study selection flow

3.2 Description of studies

The ITT outcomes were reported in 66% of the studies. The mean treatment arm size was 34.40 participants (median = 28, SD = 28.22). In total, 3543 participants were randomly assigned to an active psychotherapy, with 2295 randomly assigned to a control (eg waitlist, active supportive therapy, TAU or psychoeducation). See Table 1 for characteristics of included studies.

TABLE 1. Characteristics of included studies Study referencea Intervention No. randomised Trauma Index Diagnostic Criteria Mean age (SD) % Female Scale Acarturk et al. (2016) EMDR 49 Refugee DSM-IV 33.32 (11.09) 79.20 Impact of Event Scale-Revised (IES-R)51 WL 49 34.03 (10.00) 68.70 Aldahadha et al. (2012) EMDR 25 MVA DSM-IV-TR 26.41 (range 19–37) 54 Post-traumatic Stress Scale52 WL 26 Asukai et al. (2010) PE 12 Mixed (physical, sexual, accident) DSM-IV 27.1 (5.4) 91.66 IES-R51 TAU 12 31.4 (8.8) 83.33 Basoglu et al. (2005) CBT 31 Natural disaster DSM-IV 36.3 (11.5) 84.70 Clinician- Administered PTSD Scale (CAPS)53 WL 28 Bryant et al. (2019) CBT 33 Emergency services DSV-IV 44.7 (10.7) 12 CAPS53 WL 34 43.4 (7.8) 29 Buhmann et al. (2016) CBT 70 Refugee DSM-5 46 (8) 42 HTQ54 WL 68 47 (8) 29 Butollo et al. (2016) CPT 74 Mixed (interpersonal, accident, other) DSM-IV 37.99 (12.1) 64.90 IES-R51 DET 67 33.67 (10.3) 67.20 Capezzani et al. (2013) EMDR 11 Cancer DSM-IV 50.82 (7.74) 100 IES-R51 CBT 10 52.70 (8.68) 80 Carletto et al. (2016) EMDR 25 Health-related DSM-IV 39.52 (11.68) 75 IES-R51 ACTST 25 40.66 (10.03) 86.36 Carlson et al. (1998) EMDR 12 Military DSM-IV 45.4 (3.5) 0 Impact of Event Scale (IES)55 WL 10 52.7 (8.6) 0 Cloitre et al. (2002) STAIR-PE 31 Physical or sexual DSM-IV 34 (7.22) 100 Modified Post-traumatic Stress Disorder Symptom Scale (MPSS-SR)53 WL 27 100 Coffey et al. (2006) PE 8 Sexual and Physical DSM-IV 37.6 (8.0) 67 IES-R51 ACTST 7 Coffey et al. (2016) PE 45 Mixed DSM-IV-TR 34.7 (range 31.6–37.8) 42 IES-R51 PSYED 41 32.9 (range 29.5–36.2) 51 Cottraux et al. (2008) CBT 31 Mixed (Car accident, physical assault, rape, witness to extreme violence, incest, family violence, witness death, surgery, miscellaneous) DSM-IV 43.18 (10.6) 74.19 PCL56 ACTST 29 37.20 (9.2) 31.03 Devilly & Spence (1998) EMDR 19 Military DSM-III-R 50.1 (6.48) 0 Mississippi PTSD Scale57 TAU 16 0 Devilly et al. (1999) CBT 15 Mixed (accident, disaster, physical, sexual, war) DSM-IV 35.92 (14.53) 58.33 IES55 EMDR 17 40.18 (10.90) 72.73 Duffy et al. (2007) CT 29 civil conflict in Northern Ireland DSM-IV 44.1 (11.3) 34 Post-traumatic Diagnostic Scale (PDS)58 WL 29 43.7 (12.3) 45 Dunne et al. (2012) CBT 13 MVA DSM-IV 32.54 (7.09) 50 IES-R51 WL 13 Ehlers et al. (2005) CT 14 Mixed (accident, physical, witness death) DSM-IV 35.4 (10.9) 57 PDS58 WL 14 37.8 (11.2) 43 Ehlers et al. (2003) CT 28 MVA DSM-IV – – PDS58 PSYED 28 – – WL 29 – – Ehlers et al. (2014) CT 31 Mixed (interpersonal violence, accident/disaster, Death/harm to others/other) DSM-IV 41.5 (11.7) 58.10 PDS58 ACTST 30 37.8 (9.9) 56.70 WL 30 36.8 (10.5) 60 Fecteau & Nicki (1999) CBT 13 MVA DSM-IV 41.3 (range 25–63) 70 IES55 WL 11 Feske (2008) PE 9 Sexual and Physical DSM-IV 43.1 (range 29–55) 100 PTSD Diagnostic Scale–Interview (PDS-I)