The study was approved by the Medical Ethical Board of the St. Elisabeth hospital in Tilburg, the Netherlands and was registered in the Dutch Trial Register (www.trialregister.nl, NTR 3134). All included patients gave their written consent before enrollment. This study is a two-arm parallel RCT, including CBT and EMDR therapy.

Patients were recruited, assessed, and treated at the department of psychiatry, St. Elisabeth hospital, Tilburg, the Netherlands between February 2010 and December 2013. Advertisements were placed in a local newspaper to inform people about the existence of our study. When someone wanted to participate he or she had to visit his or her general practitioner. Patients were referred to the hospital by general practitioners.

Inclusion criteria were: (1) age between 18 and 65 years old; (2) the presence of a SCID-I primary diagnosis of PD (First et al., 1997); and (3) sufficient knowledge of the Dutch language.

Exclusion criteria were: (1) comorbid diagnosis of dementia, psychosis, severe depression, bipolar disorder, and/or another psychiatric disorder that was more prominent than the PD; (2) use of more than 20 standard units of alcohol a week; and (3) use of benzodiazepines and/or other sedative agents (De Jongh and ten Broeke, 2006). This last criterion was added because benzodiazepines or other sedative agents are likely to interfere with the level of arousal that is needed for EMDR therapy to be effective (Little et al., 2017). Patients who use modern antidepressants (e.g., Selective Serotonin Reuptake Inhibitors (SSRIs) or Serotonin and Norepinephrine Reuptake Inhibitors (SNRIs) and/or classic antidepressants (e.g., Tricyclic Antidepressants (TCA) were required to be on a stable medication dose (i.e., unchanged dosage of medication), 6 weeks prior to trial until the end. Patients were not allowed to attend any form of therapy during the whole trial. Patients not eligible for participation were offered treatment as usual.

The primary outcome measure was the severity of the PD, assessed with the Agoraphobic Cognitions Questionnaire (ACQ), which measures the degree of catastrophic cognitions when feeling anxious or tense (Chambless et al., 1984). The two subscales have a good internal consistency. The discriminant validity and construct validity are also good (Chambless et al., 1985).

The Body Sensations Questionnaire (BSQ) measures anxiety about bodily sensations and consists of two questionnaires; while the BSQ1 assesses the amount of fear, the BSQ2 measures how often the sensations are experienced when the patient feels anxious or tense (Chambless et al., 1984). The internal consistency and the test-retest reliability of the BSQ are good. Furthermore, the BSQ has good discriminant- and construct validity (Chambless et al., 1985).

The Mobility Inventory (MI) measures the degree to which places or situations are avoided with a trusted companion (MI-ac) and when the patient is alone (MI-al) (Chambless et al., 1985). Both subscales have a good internal consistency, discriminant validity and construct validity (Chambless et al., 1985). For ACQ, BCQ, and MI, lower scores indicate better outcomes.

The secondary outcome measure, QOL, was assessed with the World Health Organization Quality of Life short version (WHOQOL-Bref) (De Vries and van Heck, 1995). This measure consists of one generic facet (Overall quality of life and general health) and four domains (i.e., “Physical health,” “Psychological health,” “Social relationships,” and “Environment”) (De Vries and van Heck, 1995). Higher scores indicate better QOL. The WHOQOL-Bref is sensitive for changes over time and for treatment influences. The psychometric properties of the WHOQOL-Bref are also good (Trompenaars et al., 2005).

All patients were first screened by a psychiatrist who conducted a regular psychiatric interview, including the registration of the participants' medical status and medication use. Participation was voluntary and patients could withdraw from the study at any time without specifying a reason. After referral by a psychiatrist and before randomization, patients were screened with the Structured Clinical Interview for DSM-IV Axis I disorders (SCID-I) (First et al., 1997). The SCID-I was conducted by independent clinicians who were trained intensively during a 2-day workshop.

Patients eligible for participation were randomized to one of two treatment groups. Randomization was carried out by an independent secretary, who had 84 sealed envelopes, of which 42 contained a note with “EMDR” written on it, and 42 included a note with “CBT” on it. In both groups, a standardized treatment protocol was used. For each eligible patient, random assignment of sealed envelopes was performed. Before randomization, patients signed an informed consent. Patients were measured at baseline (T1), post-treatment (T2), and 3 months follow-up (T3), and received no financial compensation for participation.

In total, six licensed clinical psychologists (three men, three women) performed the EMDR and CBT treatments. In both groups, standardized treatment protocols were used. Therapists who performed EMDR therapy (one man, one woman) were both accredited practitioners by the European association. Therapists performing CBT treatment (three men, two women) were accredited CBT therapists by the Dutch National CBT Association.

The CBT protocol is the Dutch version of Craske and Barlow's (2008) and consists of 13 weekly sessions lasting about 60 min each (Craske and Barlow, 2008). During the first part (psycho-education), the patient is informed about panic attacks and PD. The second part consists of teaching and applying relaxation exercises which help the patient to reduce general anxiety. The third part consists of interoceptive exposure exercises in order to become accustomed to, and to cope with, the fear of bodily sensations. The fourth part is cognitive therapy in which the patients learn to recognize their automatic, anxious thoughts and formulate alternative, more adaptive thoughts. Finally, in vivo exposure consisted of learning patients to cope with the anxiety experienced during situations or activities that are feared and avoided by using an anxiety hierarchy (Kampman et al., 2004).

The EMDR treatment protocol is the Dutch version (De Jongh and ten Broeke, 2006) of Shapiro's EMDR protocol (Shapiro, 2001) and consists of 13 weekly sessions lasting about 60 min each. In this protocol, a patient is first informed about EMDR therapy, traumatic memories are identified, and the course of current symptoms is evaluated. In the present study the case conceptualization was conducted according to the “first method” of the “Two Method Approach” that deals with symptoms whereby memories of the etiological and/or aggravating events were meaningfully specified on a time line. To this end, the memories of the distressing events that were assumed to play a key role in the acquisition and maintenance of the condition and evoked distress, were determined. Subsequently, the memories that evoked the most disturbance, e.g., the first or worst panic attack, were reprocessed first using working memory taxation by listening to alternating audio tones. Subsequently, other memories that were considered to contribute to a patient's current symptoms were targeted in the same way (De Jongh et al., 2010). During EMDR therapy, patients are asked to report what associations come to mind and the patient is guided to refocus on that association. This is continued until the patient no longer reports any distress related to the target image. Afterwards, the patient is asked to formulate a positive belief regarding the target image.

To each treatment group, 20 h of group supervision by an independent qualified EMDR or CBT supervisor were given. Additional supervision by telephone or e-mail was provided on request. All patients were asked permission to make video recordings of the treatment sessions, to ensure that therapists adhered to the treatment protocol. During the study, therapists had supervision sessions in which adherence to the therapist protocol was evaluated and discussed to maintain quality and homogeneity of the intervention protocol.

According to the method of Faul et al. (2009), a sample size calculation was performed using G-Power 3.1.7 which showed that in total, 102 patients would be needed (non-inferiority test, effect size Cohen's d = 0.5, one-sided alpha = 0.05, power = 0.80). Anticipating 20% drop out, 128 patients were needed. For each outcome variable, linear mixed models (with ML estimation) were specified including main effects of group, time (categorical), and interaction effect group^*time. The dependence of the repeated measures was taken into account by using the unstructured error covariance pattern model. Covariates (i.e., age, gender, education, marital status, duration of complaints, number of axis I diagnoses, received previous treatment, and antidepressant treatment) were added to obtain adjusted results under the missing at random assumption.

Non-inferiority testing was used to determine whether EMDR is not worse than CBT (Piaggio et al., 2006, 2012). For ACQ, BCQ, and MI, the null hypothesis is (EMDR - CBT) > δ, and the alternative hypothesis is (EMDR - CBT) ≤ δ, where δ is the margin that is set at minimal clinical relevance. If the upper bound of confidence interval of 90% is below δ, it is concluded that EMDR is non-inferior to CBT. For ACQ and BCQ, the margin was δ = 5, and for MI, the margin was δ = 8. The margins of these questionnaires were determined by clinical experts. For WHOQOL-BREF, the non-inferiority was reversed and the margin was δ = −1 (Den Oudsten et al., 2013).

Group differences were analyzed at T3. Intention-to-treat approach was used on the patients that started treatment, while per-protocol approach was used as a sensitivity analysis on patients that completed all treatments (Piaggio et al., 2006, 2012). For effect size measure, Cohen's d was computed as mean difference divided by baseline pooled standard deviation. Statistical analyses were performed in SPSS version 19.0.

Information on observed outcome means and effect sizes for both treatment groups for all time points, are presented in Table 2. The intention-to-treat analyses at T3 were performed on 39 EMDR patients and 38 CBT patients. Scores on questionnaires measuring severity of PD (ACQ, BSQ1, BSQ2, MI-ac, and MI-al) showed non-inferiority of EMDR to CBT in the unadjusted analysis (Table 3, Figure 2A). In the adjusted analysis, this was also the case for ACQ, BSQ1, and BSQ2, whereas MI-ac and MI-al were inconclusive.

For the facet ‘Overall QOL and general health’ and the four QOL domains, EMDR appeared to be non-inferior to CBT at T3 in the unadjusted analysis (Table 3, Figure 2B). For the adjusted analyses, only “physical health” was inconclusive.

Per-protocol analyses included the 62 patients that had T1 and T2 measurement (10 patients were removed in the CBT group and five in the EMDR group). All conclusions were similar to the intention-to-treat analyses, except for QOL domain “Environment” in which the unadjusted analysis at T3 was now inconclusive (lower bound −1.09).

AdJ reported receiving income for published books or book chapters on EMDR and for training professionals in this method. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.