Eye movement desensitization and reprocessing (EMDR) is a trauma-focused (TF) psychotherapy developed for treating posttraumatic stress disorder (PTSD) that was first introduced in the 1980s (Shapiro, Reference Shapiro1989). This treatment involves the patient focusing on the most distressing mental images of the event while performing bilateral stimulation. After bringing up their most distressing mental images (exposure to the traumatic event), the patient's emotional arousal is interrupted by employing another stimulus (bilateral stimulation and interruption of attention) which is assumed to lead to a reduction in arousal and distress (Jeffries & Davis, Reference Jeffries and Davis2013; Rothbaum, Astin, & Marsteller, Reference Rothbaum, Astin and Marsteller2005).
Since EMDR's introduction, several mechanistic hypotheses have been proposed to explain the effects of bilateral stimulation in EMDR (Landin-Romero, Moreno-Alcazar, Pagani, & Amann, Reference Landin-Romero, Moreno-Alcazar, Pagani and Amann2018). One of them is the adaptive information processing (AIP) model (Shapiro & Laliotis, Reference Shapiro and Laliotis2011). This suggests that EMDR involves a re-setting of the system that processes and stores events during stressful situations, which reduces distress and negative emotions triggered by traumatic experiences (Shapiro, Reference Shapiro2001, Reference Shapiro2006). Another theory to explain the effects of bilateral stimulation is the working memory theory, which proposes that by taxing the working memory, eye movements permanently reduce the vividness and emotionality of aversive memories (van den Hout & Engelhard, Reference van den Hout and Engelhard2012). This theory has been supported by laboratory studies in healthy individuals (Mertens, Lund, & Engelhard, Reference Mertens, Lund and Engelhard2021). Finally, the orientating response model suggests that bilateral stimulation triggers an investigatory reflex, reducing negative emotions and enhancing awareness, facilitating exploratory behavior, and potentially improving cognitive processes (Barrowcliff, Gray, Freeman, & MacCulloch, Reference Barrowcliff, Gray, Freeman and MacCulloch2004). However, a recent meta-analysis that included dismantling studies comparing EMDR with and without eye movements, found no benefit of eye movements, casting doubt about the superiority of EMDR to trauma-focused treatments without eye movements, such as exposure therapy or cognitive behavioral therapy with a trauma-focus (CBT-TF; Cuijpers, Veen, Sijbrandij, Yoder, & Cristea, Reference Cuijpers, Veen, Sijbrandij, Yoder and Cristea2020).
Most international guidelines for the treatment of PTSD recommend the use of either CBT-TF or EMDR as first-line treatments for adults with PTSD (Hamblen et al., Reference Hamblen, Norman, Sonis, Phelps, Bisson, Nunes and Schnurr2019; Phelps et al., Reference Phelps, Lethbridge, Brennan, Bryant, Burns, Cooper and Silove2022; VA/DoD Clinical Practice Guideline, 2023). These guidelines are based on evidence-based research and provide recommendations to optimize patient treatment. Past meta-analyses (MA) have found EMDR to significantly improve PTSD symptom severity at post-test assessment (Bisson & Olff, Reference Bisson and Olff2021; Cuijpers et al., Reference Cuijpers, Veen, Sijbrandij, Yoder and Cristea2020; Cusack et al., Reference Cusack, Jonas, Forneris, Wines, Sonis, Middleton and Gaynes2016; Lewis, Roberts, Andrew, Starling, & Bisson, Reference Lewis, Roberts, Andrew, Starling and Bisson2020a). Consistent with other disorders (Papola et al., Reference Papola, Ostuzzi, Tedeschi, Gastaldon, Purgato, Del Giovane and Barbui2022), large effect sizes have been reported when comparing EMDR for PTSD to wait-list-control (WLC) groups, and smaller effects when compared to treatment-as-usual groups and other active treatment groups. While the effectiveness of EMDR against WLC has been established, it is unclear what the additive benefit of the eye movements are. Some researchers argue that the eye movements are unnecessary, while others argue that they have an added advantage (van den Hout & Engelhard, Reference van den Hout and Engelhard2012).
Currently, very little is known about moderators and predictors of EMDR treatment outcomes. Age, gender, baseline severity of PTSD, depression, and anxiety were not significantly associated with PTSD symptoms after EMDR treatment (Capezzani et al., Reference Capezzani, Ostacoli, Cavallo, Carletto, Fernandez, Solomon and Cantelmi2013). Similarly, gender did not significantly influence treatment effects in a later study (Ter Heid, Mooren, van de Schoot, de Jongh, & Keber, Reference Ter Heid, Mooren, van de Schoot, de Jongh and Keber2016). In the same study, participants who did not have refugee status had a greater reduction in PTSD symptoms compared to those with refugee status (Ter Heid et al., Reference Ter Heid, Mooren, van de Schoot, de Jongh and Keber2016). There is literature suggesting that veterans with PTSD respond less to trauma-focused treatments in general, and to EMDR specifically (Haagen, Ter Heide, Mooren, Knipscheer, & Kleber, Reference Haagen, Ter Heide, Mooren, Knipscheer and Kleber2017).
However, the methodological quality of the studies may play a role in these comparisons. There is also a great deal of inconsistency in the literature concerning the influence of specific moderators and predictors of psychotherapy outcomes in PTSD in general. No significant associations with treatment outcomes have been found for factors such as age (Ivarsson et al., Reference Ivarsson, Blom, Hesser, Carlbring, Enderby, Nordberg and Andersson2014; Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007; Lewis et al., Reference Lewis, Farewell, Groves, Kitchiner, Roberts, Vick and Bisson2017), gender (Blanchard et al., Reference Blanchard, Hickling, Devineni, Veazey, Galovski, Mundy and Buckley2003; Galovski, Blain, Mott, Elwood, & Houle, Reference Galovski, Blain, Mott, Elwood and Houle2012; Haagen et al., Reference Haagen, Ter Heide, Mooren, Knipscheer and Kleber2017; Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007; Lewis et al., Reference Lewis, Farewell, Groves, Kitchiner, Roberts, Vick and Bisson2017), marital status (Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007), employment status (Ivarsson et al., Reference Ivarsson, Blom, Hesser, Carlbring, Enderby, Nordberg and Andersson2014; Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007), therapy type (Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007), time since trauma (Ehlers et al., Reference Ehlers, Clark, Hackmann, McManus, Fennell, Herbert and Mayou2003; Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007), type of trauma (Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007), and psychiatric comorbidity (Cloitre, Koenen, Cohen, & Han, Reference Cloitre, Koenen, Cohen and Han2002; Rizvi, Vogt, & Resick, Reference Rizvi, Vogt and Resick2009).
Some individual psychotherapy studies have found that higher education (Lewis et al., Reference Lewis, Farewell, Groves, Kitchiner, Roberts, Vick and Bisson2017), higher levels of guilt (Rizvi et al., Reference Rizvi, Vogt and Resick2009), and therapeutic alliance (Cloitre et al., Reference Cloitre, Koenen, Cohen and Han2002) were associated with a better PTSD treatment response. Additionally, there is some evidence suggesting that comorbid psychiatric disorders reduce the beneficial effects of treatment on PTSD outcomes (Hagenaars, van Minnen, & Hoogduin, Reference Hagenaars, van Minnen and Hoogduin2010; van Minnen, Wessel, Dijkstra, & Roelofs, Reference van Minnen, Wessel, Dijkstra and Roelofs2002).
A long-standing issue in the field is that randomized controlled trials (RCT) and study-level (also known as aggregate or traditional) meta-analysis (MA) often lack sufficient statistical power to identify significant moderators of treatment effect (Gurung, Ellard, Mistry, Patel, & Underwood, Reference Gurung, Ellard, Mistry, Patel and Underwood2015). This may be the reason for the gap in the current literature when it comes to moderators and predictors of EMDR. An individual participant data meta-analysis (IPDMA) synthesizes raw participant-level data from multiple related studies to answer a specific set of research questions. This can be done using a one-stage or two-stage approach (Riley, Tierney, & Stewart, Reference Riley, Tierney and Stewart2021). The simpler and more utilized two-stage approach uses the participant-level data to calculate aggregate data in each trial separately, and then combines the aggregate data in a univariate MA model. This is similar to a study-level MA. On the other hand, the one-stage approach analyses participant-level data from all the trials in a single step using a generalized linear mixed model that accounts for the clustering of participants within trials. These two approaches tend to give very similar results when the same assumptions and estimation methods are used. However, when the number of trials included in the IPDMA is small then a one-stage approach is more exact (Riley et al., Reference Riley, Tierney and Stewart2021). Using an IPDMA approach, we can maximize statistical power to detect more precise effects and explore participant-level characteristics as moderators of treatment outcomes.
By gaining insight into potential moderators and predictors of the effectiveness of EMDR, we may have better precision to identify patients who would benefit the most from EMDR. This is important since EMDR is highly protocolized, relatively straightforward to administer, and requires shorter episodes of imaginal trauma exposure (Nijdam, Gersons, Reitsma, de Jongh, & Olff, Reference Nijdam, Gersons, Reitsma, de Jongh and Olff2012; Schubert & Lee, Reference Schubert and Lee2009) in comparison to CBT-TF psychotherapies. The aims of this study were to (1) investigate the effectiveness of EMDR in reducing PTSD symptom severity, achieving treatment response, attaining PTSD remission, and reducing treatment dropout rates in comparison to another psychological treatment among adults with PTSD, and (2) explore potential sociodemographic, clinical, and intervention-related moderators of EMDR treatment effects among adults with PTSD.
Methods
This IPDMA was reported in compliance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) IPD Statement (Stewart et al., Reference Stewart, Clarke, Rovers, Riley, Simmonds and Stewart2015). We registered the study on PROSPERO (CRD42020138638). Additional information can also be found in the protocol paper (Wright et al., Reference Wright, Karyotaki, Bisson, Cuijpers, Papola, Witteveen and Sijbrandij2022).
Eligibility criteria
Study inclusion was limited to RCTs comparing EMDR to active treatments. Active treatments included other psychological treatments (e.g. relaxation therapy, emotional freedom technique, and trauma-focused cognitive behavioral psychotherapies). We excluded studies comparing EMDR to WLC and treatment-as-usual groups. Studies were included if participants were adults (18 years or older) with above-threshold PTSD symptoms based on any established self-report scale or a clinical diagnosis of PTSD. Within each study, at least 70% of the participants were required to have a clinical diagnosis of PTSD according to any version of the DSM or ICD criteria. Only studies published in English were included.
Study identification and selection process
We used an existing database of psychological treatments for PTSD that was created by the Cardiff University Traumatic Stress Research Group (CUTSRG) to perform a systematic review and MA for the treatment guidelines of the International Society for Traumatic Stress Studies (ISTSS). A search was conducted by the Cochrane Collaboration Cochrane with the same inclusion criteria including all studies published until May 2018 (Bisson, Andrew, Roberts, Cooper, & Lewis, Reference Bisson, Andrew, Roberts, Cooper and Lewis2013; Bisson et al., Reference Bisson, Berliner, Cloitre, Forbes, Jensen, Lewis and Shapiro2019; Lewis et al., Reference Lewis, Roberts, Andrew, Starling and Bisson2020a). We updated the search using the same search strategy (see online Appendix 1). The search strategy included the screening of major bibliographic databases such as PubMED, Embase, PsycINFO, PTSDpubs, and CENTRAL. We also screened past systematic reviews for additional articles. Our updated searches included studies published between the May 1, 2018 and the 11 January 2021 (see online Appendix 2).
Two researchers (SLW and DP) independently screened the titles and abstracts for the initial update (1 May 2018 till the 13 May 2019). Titles and abstracts for the second update (1 January 2019 till the 11 January 2021) were also screened by two researchers (SLW and ABW). The same researchers screened the full text of studies that possibly met the inclusion criteria. Senior members of the review team (MS and GS) resolved any uncertainties regarding study inclusion.
Data collection
Authors of eligible trials were contacted to request the use of their participant-level datasets. At least six additional reminder emails were sent at intervals of two to three weeks if no response was received. If a response had not been received at this point, an additional two authors were contacted (when possible). If after this no response was received, we attempted to reach out to the corresponding authors on ResearchGate and LinkedIn. If a response was still not received, the dataset was considered unavailable along with those where the corresponding author declined to share data. In cases where the author was able to provide their participant-level dataset, data collection and storage were conducted in alignment with the European General Data Protection Regulation (Regulation (EU) 2016/679). Participant-level datasets were anonymized before sharing and were stored in an existing encrypted, password-protected folder at Vrije Universiteit Amsterdam. For data protection purposes, we sent the link for the encrypted folder to a different email address than the one we used for the password. Our data collection commenced at the end of 2019 as planned, however it was extended to two and a half years because an insufficient number of datasets had been collected by the end of 2020.
Primary and secondary outcomes
The primary outcome data included PTSD symptom severity, treatment response, and PTSD remission. The secondary outcome was treatment dropout. PTSD symptom severity scores were measured using PTSD measures and clinical interviews. Studies included in this IPDMA used self-report measures like the Impact of Event Scale (total score ranging from 0 to 75; Horowitz, Wilner, & Alvarez, Reference Horowitz, Wilner and Alvarez1979), Impact of Event Scale-Revised (IES-R; total score ranging from 0 to 88; Weiss & Marmar, Reference Weiss, Marmar, Wilson and Keane1997), PTSD Checklist (PCL; total score ranging from 0 to 80; Blanchard, Jones-Alexander, Buckley, & Forneris, Reference Blanchard, Jones-Alexander, Buckley and Forneris1996), and the Mississippi Scale for PTSD (MPTSD; total score ranging from 35 to 175; Keane, Malloy, & Fairbank, Reference Keane, Malloy and Fairbank1984). We used self-report measures because we wanted to use the same type of assessment across studies. All available studies provided self-report PTSD severity outcomes but not all studies provided PTSD severity outcome data based on clinical interviews. Treatment response was defined as a 50% reduction in baseline PTSD symptom scores (Karyotaki et al., Reference Karyotaki, Riper, Twisk, Hoogendoorn, Kleiboer, Mira and Cuijpers2017). Participants were considered in PTSD remission if they no longer had a formal diagnosis of PTSD at the post-test. In line with the definition used by CUTSRG for ISTSS' treatment guidelines, we considered a participant who left the study before the post-test assessment as a treatment dropout (Lewis, Roberts, Gibson, & Bisson, Reference Lewis, Roberts, Gibson and Bisson2020b). Assessment time points included baseline and post-test. In addition, we requested available baseline sociodemographic, clinical, and intervention-related variables in order to investigate their effect on treatment outcomes (when there was sufficient data to do so). Moderators included gender; age; relationship status; partner or no partner (married/cohabitating or divorced/widowed/separated/single); marital status (yes or no); tertiary education (yes or no); employed (yes or no); using psychotropic medication (yes/no); above threshold depression (yes/no); anxiety threshold (yes or no); chronic (duration of PTSD symptoms > 3 months; yes or no), and comorbid psychiatric disorder (yes or no).
Risk of bias assessment
Studies were assessed for risk of bias using Cochrane Risk of Bias 2 Tool (RoB 2; Higgins, Savović, Page, Elbers, & Sterne, Reference Higgins, Savović, Page, Elbers, Sterne, Higgins, Thomas, Chandler, Cumpston, Li, Page and Welch2022). Each study was assessed independently by two researchers (SW and SS) to determine whether there was a risk for bias related to bias arising from the following domains: randomization process (D1), deviations from intended intervention (D2), and measurement of the outcome (D4). We did not include domains 3 and 5 (missing outcome data and selection of the reported result respectively) because missing values were addressed in our analyses, and bias related to the selection of reported outcomes because we had access to the participant-level datasets (Karyotaki et al., Reference Karyotaki, Sijbrandij, Purgato, Acarturk, Lakin, Bailey and Barbui2023). Each domain comprises signaling questions which lead to a domain-level judgement on the risk of bias. All uncertainties were cleared up through deliberation by SW and SS with the assistance of an additional member of the review team.
Data analysis
Data analysis was conducted in Stata 17 (StataCorp, 2021). We combined all individual data sets into a merged data set, using a generic standardized protocol for integrating participant-level datasets (Stewart et al., Reference Stewart, Clarke, Rovers, Riley, Simmonds and Stewart2015). Study-level variables were included for both the available and unavailable participant-level datasets, which were extracted from the studies' available documentation, such as publications and trial registries.
Study-level MA
We conducted a study-level MA to examine the differences across the studies that provided participant-level data and the studies for which participant-level data were unavailable. The difference between the studies that did and did not provide data was compared in a subgroup analysis. Heterogeneity was examined by calculating I 2 indicating heterogeneity as a percentage with 25% as low, 50% as moderate, and 75% as high (Higgins & Thompson, Reference Higgins and Thompson2002). The 95% confidence intervals (CI) around I 2 were calculated using the non-central chi-squared-based approach in the heterogi module of Stata (Ioannidis, Patsopoulos, & Evangelou, Reference Ioannidis, Patsopoulos and Evangelou2007; Orsini, Bottai, Higgins, & Buchan, Reference Orsini, Bottai, Higgins and Buchan2005). We examined publication bias by visually inspecting the funnel plot, using the trim and fill procedure and Egger's test of funnel plot asymmetry (Duval & Tweedie, Reference Duval and Tweedie2000; Egger, Davey Smith, Schneider, & Minder, Reference Egger, Davey Smith, Schneider and Minder1997).
IPDMA
PTSD total scores were standardized by transforming each study's PTSD severity original (raw) scores to z scores before combining the individual datasets into the master dataset. Missing outcome data for PTSD symptom severity at post-test were estimated using multiple imputation under the missing-at-random assumption (miimputemvnin STATAsoftware, version 17; StataCorp, 2021). This method generated 20 imputed data sets using data on baseline PTSD symptom severity scores. These newly imputed datasets included the observed and the imputed standardized PTSD symptom scores for missing values. They were analyzed separately using the selected model, and the results were averaged according to Rubin's rules (Riley, Lambert, & Abo-Zaid, Reference Riley, Lambert and Abo-Zaid2010).
One-stage IPDMA
In a one-stage IPDMA, we merged all participant-level data from all studies with participants clustered within studies. One-stage IPDMA yields more precise and less biased estimates of effect, maximizes the statistical power, and accounts for parameter correlation (Debray, Moons, Abo-Zaid, Koffijberg, & Riley, Reference Debray, Moons, Abo-Zaid, Koffijberg and Riley2013; Stewart & Parmar, Reference Stewart and Parmar1993). We calculated the standardized β coefficient for the examined comparisons. This estimate indicates how many s.d.s the dependent variable changes per s.d. increase in the predictor variable. Thus, the higher the β the greater the effect of the predictor variable on the dependent variable, although there is no association among the variables if the β is 0. Using a one-stage approach, we analyzed the effect of the interventions on PTSD symptom severity at the end of treatment with a multilevel mixed-effects linear regression using a random intercepts model with a random effect for each trial and fixed effects for the intervention and symptom severity, using STATA's mixed command. Post-test PTSD scores were used as the dependent variable and trial arm condition (EMDR v. control) as the independent variable while controlling for baseline PTSD symptom severity. We analyzed the effect of the interventions on treatment response at post-test using a multilevel mixed-effects logistic regression (using a random intercepts model with a random effect for each trial and fixed effects for the intervention and PTSD symptom severity, using STATA's melogit command). Response (yes or no) was the dependent variable, and treatment condition was the independent variable. This was repeated for PTSD remission and dropout.
Two-stage IPDMA
The two-stage approach uses participant-level data to derive aggregate data (such as treatment effect estimates) in each trial separately, and then combines the aggregate data in a study-level MA model. We ran a two-stage IPDMA using STATA's ipdmetan command.
Moderator analyses
We tested whether available demographic and clinical characteristics (gender, age, relationship status, marital status, completed some form of tertiary education, employment status, chronic PTSD status, use of psychotropics, above threshold depression, and presence of comorbidity) moderated the effect of EMDR on PTSD outcomes. Not all included studies reported data on the selected moderators. We included moderator analyses when the variables were reported by three or more studies. To examine moderators, we added the interaction between each potential moderator and treatment outcome on PTSD severity into a multilevel mixed-effects linear regression model. We similarly added the interaction between each potential moderator and treatment response into a multilevel mixed-effects logistic regression model. Each potential moderator was included in a separate model as the main effect.
Sensitivity analysis
Sensitivity analyses were conducted to examine the effect of EMDR compared to CBT-TF. We also compared the effect of EMDR to other psychotherapies, while excluding CBT-TF.
Results
Study selection and participant-level data obtained
The systematic literature search resulted in 15 eligible articles. We were able to obtain participant-level data from eight studies, comprising 346 participants (Ahmadi, Hazrati, Ahmadizadeh, & Noohi, Reference Ahmadi, Hazrati, Ahmadizadeh and Noohi2015; Capezzani et al., Reference Capezzani, Ostacoli, Cavallo, Carletto, Fernandez, Solomon and Cantelmi2013; Carletto et al., Reference Carletto, Borghi, Bertino, Oliva, Cavallo, Hofmann and Ostacoli2016; Devilly & Spence, Reference Devilly and Spence1999; Karatzias et al., Reference Karatzias, Power, Brown, McGoldrick, Begum, Young and Adams2011; Laugharne et al., Reference Laugharne, Kullack, Lee, McGuire, Brockman, Drummond and Starkstein2016; Lee, Gavriel, Drummond, Richards, & Greenwald, Reference Lee, Gavriel, Drummond, Richards and Greenwald2002; Nijdam et al., Reference Nijdam, Gersons, Reitsma, de Jongh and Olff2012) and these were included in the analyses reported herein. Seven eligible datasets were unavailable and could not be included in this IPDMA (Boterhoven-De Haan et al., Reference Boterhoven-De Haan, Lee, Fassbinder, van Es, Menninga, Meewisse and Arntz2020; Carlson, Chemtob, Rusnak, Hedlund, & Muraoka, Reference Carlson, Chemtob, Rusnak, Hedlund and Muraoka1998; Power et al., Reference Power, McGoldrick, Brown, Buchanan, Sharp, Swanson and Karatzias2002; Rothbaum et al., Reference Rothbaum, Astin and Marsteller2005; Taylor et al., Reference Taylor, Thordarson, Maxfield, Fedoroff, Lovell and Ogrodniczuk2003; Ter Heid et al., Reference Ter Heid, Mooren, van de Schoot, de Jongh and Keber2016; Vaughan et al., Reference Vaughan, Armstrong, Gold, O'Connor, Jenneke and Tarrier1994). Of the unavailable studies, corresponding authors reported that two were lost, three did not respond to the study invitations, one indicated that consent issues precluded data sharing, and one study was still in progress. While the main results had been published (Boterhoven-De Haan et al., Reference Boterhoven-De Haan, Lee, Fassbinder, van Es, Menninga, Meewisse and Arntz2020), the authors decided not to share their data because the additional papers were still being written up. See online Appendix 3 for a summary of study characteristics for the unavailable studies.
Study and participant characteristics
The eight studies evaluated EMDR against the following treatments: one relaxation therapy (Carletto et al., Reference Carletto, Borghi, Bertino, Oliva, Cavallo, Hofmann and Ostacoli2016), one REM-Desensitization (Ahmadi et al., Reference Ahmadi, Hazrati, Ahmadizadeh and Noohi2015), one emotional freedom technique (Karatzias et al., Reference Karatzias, Power, Brown, McGoldrick, Begum, Young and Adams2011), one Prolonged Exposure (CBT-TF [PE]) (Laugharne et al., Reference Laugharne, Kullack, Lee, McGuire, Brockman, Drummond and Starkstein2016), one brief eclectic psychotherapy (CBT-TF [BEP]) (Nijdam et al., Reference Nijdam, Gersons, Reitsma, de Jongh and Olff2012), and three CBT-TF (Unspecified) (Capezzani et al., Reference Capezzani, Ostacoli, Cavallo, Carletto, Fernandez, Solomon and Cantelmi2013; Devilly & Spence, Reference Devilly and Spence1999; Lee et al., Reference Lee, Gavriel, Drummond, Richards and Greenwald2002). Both the EMDR and comparator treatments had significant improvements in overall PTSD symptom severity at the post-test assessment in comparison to their group baseline scores.
All eight EMDR studies used the standard EMDR protocol (Shapiro, Reference Shapiro1989). The included studies were conducted in the following countries: Iran (1), Netherlands (1), Australia (3), Italy (2), and Scotland (1). All the included EMDR interventions were conducted in person and in a one-on-one format (see online Appendix 4). Additionally, all participants had a diagnosis of PTSD at the baseline assessment. The mean (s.d.) age of participants was 38.61 (11.90) years. 204 (59.13%) of 345 were female, 125 (51.02%) of 245 were married or cohabitating, and 101 (41.22%) were single. 109 (53.17%) of 205 had completed some form of tertiary education, 133 (53.63%) of 248 had no comorbid diagnosis at the baseline assessment, 95 (41.67%) of 228 were stable on psychotropic medication, and 261 (95.96%) of 272 had chronic PTSD at the baseline assessment (see online Appendix 5). The mean (s.d.) baseline PTSD symptom score was 53.78 (10.63; 95% CI 50.76–56.80) on the Impact of Event Scale (Horowitz et al., Reference Horowitz, Wilner and Alvarez1979), 68.88 (20.28; 95% CI 66.07–71.69) on the Impact of Event Scale-Revised (IES-R; Weiss & Marmar, Reference Weiss, Marmar, Wilson and Keane1997); 57.61 (10.76; 95% CI 54.96–60.25) on the PCL (Blanchard et al., Reference Blanchard, Jones-Alexander, Buckley and Forneris1996), and 148.43 (11.23; 95% CI 143.31–153.54) on Keane's Post-Traumatic Stress Disorder Scale from the Minnesota Multiphasic Personality Inventory (MMPI-K; Keane et al., Reference Keane, Malloy and Fairbank1984) in the respective studies. Finally, 94 (33.22%) out of the 283 participants dropped out of treatment before the post-test assessment.
Risk of bias assessment
Online Appendix 6 presents the RoB2 ratings for the studies included in this IPDMA. One of the eight studies scored some concerns on D1, risk of bias arising from the randomization process because the information about the randomization methods was limited to a statement that the study was randomized. Additionally, masking participants is difficult to achieve in psychotherapy research, which resulted in three studies being rated as having some concerns. None of the included studies were rated as being at a high risk of bias on any of the domains.
Results of study-level MA
Fifteen studies compared EMDR with another psychological treatment. The results of the study-level MA of all 15 included studies revealed no significant difference in PTSD symptom severity between EMDR and the comparator interventions at the post-test assessment g = −0.091, 95% CI −0.33 to 0.15, p = 0.462. Heterogeneity was moderate, I 2 = 55.79%. There was no significant difference between the outcome findings of studies included in the present IPDMA and studies with unavailable data, p = 0.87. See online Appendix 7. Based on a visual inspection of the funnel plot of standard error Hedges' g, it is unlikely that publication bias is present in this MA (see online Appendix 8).
Primary outcomes
IPDMA: PTSD symptom severity
Online Appendix 9 presents the main and moderator results of a one-stage IPDMA on PTSD symptom severity at the post-test. A one-stage IPDMA found no significant difference between EMDR and comparator interventions on PTSD symptom severity, β = −0.24, 95% CI −0.62 to 0.14, p = 0.210, n (studies) = 270 (8) in the completer analysis. The full sample one-stage IPDMA analysis based on imputed PTSD severity outcome data, β = −0.20, 95% CI −0.52 to 0.12, p = 0.217, n (studies) = 339 (8), and the two-stage yielded a similar result to the one-stage IPDMA completer analysis, g = –0.20, 95% CI −0.55 to 0.14, p = 0.251.
Baseline PTSD symptom severity was found to be a significant predictor of post-test PTSD severity in the one-stage completer analysis (β = 0.43; p = 0.000), and the imputed full sample analysis, β = 0.43, p = 0.000. More specifically, higher baseline PTSD symptom severity was associated with higher post-test PTSD severity.
In the completer analysis only, employment status significantly moderated the relationship between therapy type and post-test PTSD symptom severity. More specifically, unemployed participants who received EMDR reported significantly higher PTSD symptom severity at the post-test than employed participants who received EMDR, β = 0.80, p = 0.019. None of the other participant-level variables (sociodemographic, clinical, and intervention-related characteristics) significantly moderated PTSD symptom severity after EMDR treatment in the completer or imputed analyses (see online Appendix 9).
IPDMA: treatment response
In one-stage analysis, no significant difference in the effect of EMDR compared with other psychological treatments was found for PTSD treatment response, β = 0.86, 95% CI −0.03 to 1.74, p = 0.057, n (studies) = 270(8). See online Appendix 10. The OR was 2.36. The two-stage analysis MA also found no significant difference in effect between EMDR and other psychological treatments for PTSD treatment response, β = 0.52, 95% CI −0.42 to 1.46, p = 0.278. The OR was 1.68.
Employment status significantly moderated the relationship between therapy type and PTSD treatment response. More specifically, unemployed participants who received EMDR were significantly less likely to have responded to treatment at post-test than the employed participants who received EMDR, β = −0.63, p = 0.005.
None of the other sociodemographic, clinical, and intervention-related characteristics of participants was significantly associated with treatment response (see online Appendix 11).
IPDMA: PTSD remission
In a one-stage IPDMA, no significant difference in effect between EMDR and other psychological treatments for PTSD remission at post-test were found, β = 1.05, 95% CI −0.11 to 2.22, p = 0.075, n (studies) = 199(5). See online Appendix 11. The OR was 2.87. The two-stage analysis found a significant effect of EMDR compared with other psychological treatments for PTSD remission at post-test, g = 1.00, 95% CI 0.14–1.87, p = 0.023. The OR was 2.73. There was insufficient data to run a moderator analysis.
Secondary outcome
IPDMA: treatment dropout
One-stage IPDMA on treatment dropout found no significant difference in effect between EMDR and other psychological treatments at post-test, β = −0.25, 95% CI −0.79 to 0.29, p = 0.369, n (studies) = 283 (6). See online Appendix 12. The OR was 0.78. The two-stage analysis found no significant difference in the effect of EMDR over controls for PTSD treatment dropout at post-test, β = −0.19, 95% CI −0.83 to 0.45, p = 0.553. The OR was 0.82.
Gender significantly moderated the relationship between therapy type and PTSD dropout. More specifically, male participants in EMDR groups were significantly more likely to drop out of EMDR treatment than female participants in EMDR groups, β = 0.23, p = 0.028. None of the other sociodemographic, clinical, and intervention-related characteristics of participants was significantly associated with treatment dropout (see online Appendix 12).
Sensitivity analysis
We re-ran the one-stage IPDMA, including only the CBT-TF comparison groups. We found no significant difference in effect between EMDR and CBT-TF comparison groups, β = −0.18, 95% CI −0.75 to 0.38, p = 0.525, n (studies) = 180 (5).
We re-ran the one-stage IPDMA, excluding the CBT-TF comparison groups. We found no significant difference in effect between EMDR and the other comparison groups, β = −0.29, 95% CI −0.82 to 0.25, p = 0.294, n (studies) = 90 (3).
Discussion
To the best of our knowledge, this is the first IPDMA to explore moderators of EMDR for adults with PTSD using individual participant-level data. One of the strengths of the present study was the statistical power to detect statistically significant moderators compared with study-level MA and published RCTs aimed at investigating the efficacy of EMDR for adults with PTSD. This IPDMA made it possible to investigate available participant-level moderators, such as employment status and gender.
In line with past research, the current study found no significant difference between EMDR and other psychological treatments on PTSD severity, treatment response, or treatment dropout in either the one- or two-stage IPDMA (Lewis et al., Reference Lewis, Roberts, Andrew, Starling and Bisson2020a). Despite the proposed different mechanisms of action, no significant difference in efficacy was found between EMDR and CBT-TF comparison groups. When we removed the CBT-TF comparison groups from the analysis, there was still no significant difference. The current study did not provide support for the notion that psychotherapies with eye movements are more effective in treating PTSD than those without eye movements.
It is important to note that all the psychological comparator treatment groups were found to be effective in treating PTSD when interpreting these findings. We found no significant difference between EMDR and the psychological treatment control groups on PTSD remission at post-test in the one-stage analysis. However, a significant main effect in favor of EMDR was found in the two-stage analysis on PTSD remission. Considering the small sample sizes of the included trials, the one-stage IPDMA result is most likely the more accurate reflection of the true effect.
In line with previous research, baseline PTSD was a significant predictor of post-test PTSD symptom severity (Taylor et al., Reference Taylor, Thordarson, Maxfield, Fedoroff, Lovell and Ogrodniczuk2003). Specifically, higher baseline PTSD symptom severity was associated with higher post-test PTSD symptom severity. An earlier study found that higher PTSD baseline scores on PTSD self-report measures were associated with better treatment outcomes on self-report PTSD questionnaires (Karatzias et al., Reference Karatzias, Power, McGoldrick, Brown, Buchanan, Sharp and Swanson2007). Overall, this was not the case in this aggregated set of trials. While there are distinct differences between EMDR and the other TF therapies, our finding suggests they are equally efficacious at treating PTSD symptoms. These findings are in line with past study-level MA (Lewis et al., Reference Lewis, Roberts, Andrew, Starling and Bisson2020a).
Our moderator analysis was exploratory in nature and based on available sociodemographic, clinical, and intervention-related variables available in the obtained databases. Results from our completer moderator analysis found unemployed participants who received EMDR reported significantly higher PTSD symptom severity at post-test than employed participants who received EMDR. Similarly, we found unemployed participants who received EMDR were significantly less likely to respond to treatment by post-test than the participants who received EMDR and were employed.
Past research supports our current findings. Unemployed participants were found to be more likely to suffer from higher levels of mental health problems including PTSD (Bosmans & van der Velden, Reference Bosmans and van der Velden2018; McKee-Ryan, Song, Wanberg, & Kinicki, Reference McKee-Ryan, Song, Wanberg and Kinicki2005). In a longitudinal study among employed and unemployed trauma-exposed participants, unemployed participants continued to experience higher levels of mental health problems even years after exposure (Bosmans & van der Velden, Reference Bosmans and van der Velden2018). Research has attributed the benefits of employment as income, status, relationships, and esteem (Chen, Westman, & Hobfoll, Reference Chen, Westman and Hobfoll2015; Paul & Batinic, Reference Paul and Batinic2010). Unemployed participants might be more socially isolated in comparison to their employed counterparts or have more severe or further advanced symptoms (Nijdam, Vermetten, & McFarlane, Reference Nijdam, Vermetten and McFarlane2023) resulting in less beneficial PTSD outcomes at post-test. Furthermore, it is possible that financial concerns and associated psychological distress could distract or hinder the recovery process in unemployed participants. Another explanation for poorer outcomes for unemployed participants may be that unemployed participants may more often be engaged in compensation seeking procedures that may cause additional distress. Indeed, a direct relationship was found between compensation-related distress and higher PTSD symptom severity (O'Donnell et al., Reference O'Donnell, Grant, Alkemade, Spittal, Creamer, Silove and Studdert2015).
We also found that male participants who received EMDR were significantly more likely to drop out of treatment than female participants who received EMDR. Both brain and behavior differences in men and women may explain why men were found to be more likely to drop out of EMDR treatment in comparison to female participants (Olff, Reference Olff2017). In a recent survey among Australian males who attended mental health services, various reasons for drop-out were self-reported, among which was a lack of connection with the therapist, the sense that therapy lacked progress and the cost/inconvenience related with attending therapy sessions (Seidler et al., Reference Seidler, Wilson, Kealy, Oliffe, Ogrodniczuk and Rice2021). Thus, it is crucial that studies examine strategies to make interventions more attractive and acceptable for males, to prevent drop-out.
To our knowledge, this study is the first MA to use individual participant-level data to examine moderators of EMDR for adults with PTSD. Among the strengths of the present study was its higher power to detect statistically significant moderators compared with study-level MA or any of the current RCTs aimed at investigating the efficacy of EMDR for adults with PTSD. The use of an IPDMA made it possible to investigate participant-level moderators (such as employment status and gender). Based on a visual inspection of the funnel plot of standard error Hedges' g, it is unlikely publication bias is present in this MA.
Several limitations of our IPDMA should be mentioned. First, the small sample sizes of the included studies, and consequently the total number of participants included in this IPDMA, limited our ability to detect certain moderators. Second, our findings are at risk of availability bias because we could not access data from seven eligible studies. However, the results of the study-level MA indicated no significant difference between the studies included in the present IPDMA and studies with unavailable data. Six of the 15 studies were published more than 20 years ago (≤2003). Only two of these six older studies were available for this IPDMA (Devilly & Spence, Reference Devilly and Spence1999; Lee et al., Reference Lee, Gavriel, Drummond, Richards and Greenwald2002). Additionally, we could not examine several variables that could potentially influence EMDR treatment response, such as symptom duration or the number of treatment sessions attended because we did not have sufficient studies reporting these variables to conduct these analyses. Two studies were excluded from the dropout analysis because only the completer data were provided (Ahmadi et al., Reference Ahmadi, Hazrati, Ahmadizadeh and Noohi2015; Carletto et al., Reference Carletto, Borghi, Bertino, Oliva, Cavallo, Hofmann and Ostacoli2016). Furthermore, most of the participants had chronic PTSD. Therefore, the current findings can only be generalized to patients with chronic PTSD.
Considering that EMDR is highly protocolized, and is relatively straightforward to administer, it may be a more cost and resource-effective treatment option to implement in areas with limited human resources. In the only systematic review to compare the relative cost-effectiveness of different PTSD treatments, EMDR was found to be the most cost-effective (Mavranezouli et al., Reference Mavranezouli, Megnin-Viggars, Grey, Bhutani, Leach, Daly and Pilling2020). However, further studies are needed in this area, in particular, large international RCTs.
With the increased use of secondary analyses and IPDMA, researchers are strongly urged to anonymize and store their data (in a usable format) for long-term use. In terms of the FAIR data principles, this not only improves scientific integrity but also prevents us from overlooking important discoveries. Data sharing, compiling and storage have become much faster and easier. By increasing the sample sizes in the EMDR effectiveness trials, reducing risk of bias, and increasing the number of RCTs statistical power can be increased which could improve our precision in detecting clinically relevant moderators of treatment outcomes. An update of this IPDMA in the future may have greater statistical power to provide further insight into moderating effects of participant-level factors on PTSD treatment response. Further research on the mechanisms underlying the efficacy of EMDR is needed, as well as whether bilateral stimulation offers additional therapeutic benefit. This requires adequately powered studies with dismantling designs. It is also necessary to carry out well-designed randomized controlled trials in large samples.
In sum, this is the first IPDMA to have examined the effect of EMDR in comparison to other psychological treatments and explore what individual-level characteristics moderate PTSD treatment outcomes. Overall, no significant difference was found between EMDR and other effective psychological treatments in terms of PTSD outcomes. The use of IPDMA allowed us to examine the moderators of EMDR in a more powerful manner. It is vital that studies investigate approaches to make interventions more attractive and acceptable for males, in order to reduce drop-out. Findings from this IPDMA suggest that particular consideration should be taken when administering EMDR to unemployed individuals who may have extra challenges that could influence their treatment course.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723003446.
Funding statement
This research is funded by the NRF-NUFFIC scholarship, grant number 115977.
Competing interests
None.