Study design and population
The randomized controlled trial was conducted between July 2012 and May 2018. The patients were randomized to either a manualized CBGT condition or a comparator condition, MBSR. An active treatment comparison was chosen as the control condition [15]. We judged it unethical to withhold treatment from the control group because the sample consisted of individuals presenting problems with anger and violent behaviour towards others. Also, previous research has highlighted the need for, and importance of future studies comparing active treatments for intimate partner violence [10, 16]. All consecutive voluntary referrals by general practitioners of partner-violent adult men (N = 227) to St. Olav’s University Hospital, Forensic Department and Research Centre Brøset’s therapy program for aggressive and violent behavior, were assessed for eligibility. The inclusion criteria for participation in the study were that participants were male, aged 18 years or older, were violent towards a current partner or ex-partner, understood and spoke Norwegian fluently, admitted to having problems with anger and violence towards their female partner and provided written informed consent for study participation. Patients were excluded from joining the study if they were violent towards non-partners only, had current uncontrolled psychotic symptoms, or were using drugs or alcohol to a degree that it was impossible to be sober during treatment sessions. Those who met the necessary criteria and agreed to participate (n = 144) were randomly assigned to one of the two treatment conditions. Figure 1 describes the participant flow from recruitment to study completion. Partners of participating perpetrators were also eligible if they agreed to take part (N = 56).
Recruitment and procedures
All participants (perpetrators and partners) received written and verbal information about the study prior to recruitment. Postgraduate hospital staff conducted the consent procedures, and written consent was obtained from the participants prior to approaching partners and beginning baseline assessments. Provided the patients gave their written consent, their partners were contacted by phone to inform them about the study and to ask for their consent to participate in it. The partners were assured that their responses to the questionnaires would remain confidential. Those who agreed sent a written consent by mail to the hospital. In cases where any participant subsequently wished to remove their consent to participate, they approved inclusion of already collected anonymous data in the study.
The intervention and the comparator groups
Two active interventions were compared in this study: Cognitive behavioural group therapy (intervention group) versus mindfulness-based stress reduction group therapy (comparator group). Both interventions were delivered in an outpatient health service setting.
Cognitive behavioural group therapy (CBGT)
Participants in the intervention group received two individual sessions followed by 15 cognitive-behavioural group therapy sessions (total 30 h). The groups consisted of four to six patients in each group The key principles used in the CBGT focused on establishing a therapeutic relationship, behavioural change strategies, cognitive restructuring, modification of core beliefs and schemas, and the prevention of relapse and recurrent violence [6, 17]. The first five sessions were psychoeducational and focused on dysfunctional anger and how information processing was tied to affective, motivational and behavioural responses in humans. Information about the consequences of domestic violence on the victims (partner and children) in the family was discussed and the patients’ pattern of violence was explored by identifying typical risk situations. The remaining ten sessions all began with a review of a practice assignment (e.g. practice in communication and partner conflict resolution skills). Each patient presented a violent episode to the group, which was analysed by exploring negative automatic thoughts and maladaptive beliefs activated in the particular situation, and by reviewing the evidence for and against these thoughts and beliefs and considering alternative interpretations of the situation that led to violence. In addition, by practising on taking time-outs in violence risk situations, the patients were trained to accept and cope with negative emotions without acting them out. Toward the end of the group therapy, the patients created action plans for future conflict risk situations.
Mindfulness-based stress reduction group therapy (MBSR)
The comparator condition consisted of one individual session before and one session after eight group sessions of mindfulness-based stress reduction group therapy (16 h) [18, 19]. The aim of the comparator condition was to develop the skills of noticing the presence of negative thoughts without avoiding them. Also, it aimed to enhance consciousness of body sensations and mood in anger provoking situations, awareness of the interaction with others in high-arousal situations and learning of skills to manage negative emotions without acting them out violently. The key principles used in the MBSR focused on techniques derived from meditation and yoga to counteract stress and create a balance of body and mind. The groups size varied between four to six patients. The patients were expected to practice new skills every day at home between the sessions [18, 19].
Group therapists
Three psychiatric nurses delivered the CBGT and specialists in clinical psychology and education delivered the MBSR. Both interventions were manualized, and all five therapists had formal education and training in cognitive behaviour therapy and mindfulness respectively. Adherence to the study protocol was monitored through regular meetings between the therapists and the researchers. Furthermore, the therapists detailed the content of each group session in clinical records. A research assistant monitored the clinical documentation.
Outcome measurements
The baseline assessments were completed by the participants based on self-report, guided by the hospital staff after the randomization, but completed without assistance at the follow-up assessments. The primary outcome was assessed at baseline and at 3, 6, 9 and 12 months’ after baseline. For the follow-up assessments, the participants (perpetrators and partners) could choose to self-report electronically or by paper. The participants who did not return their questionnaires within a week were contacted by SMS, encouraging them to convey their answers. If they still did not answer, the follow-up questionnaires were posted again up to two times. The respondents were also encouraged to contact the study research assistant if they needed help to complete the questionnaires.
Primary outcome
The pre-defined primary outcome was change in violent behavior at 12 months’ follow-up. Violence was assessed over the preceding 3 months, as reported by the male participants and their female partners at baseline and at 3, 6, 9 and 12 months’ follow-up, using the Norwegian version of the revised Conflict Tactics Scales (CTS2) [20, 21]. Data was collected from the CTS2 subscales physical violence (12 items), physical injury (6 items), psychological violence (10 items), and sexual violence (3 items). The CTS2 is a widely used instrument and measures four dimensions of intimate partner violence, i.e. the extent of physical, psychological and sexual violence and resulting physical injury [22]. Cronbach’s alpha for the current sample (both conditions combined) was .89. The response categories for each item range from 0 to 7 and measure incidence over the previous 3 months (0 = never happened, 1 = happened once, 2 = happened twice, 3 = happened 3–5 times, 4 = happened 6–10 times, 5 = happened 11–20 times, 6 = happened more than 20 times). The standard CTS2 has a seventh score (7 = never happened in the last 3 months, but has happened before). In this study the seventh score was not used since it was focused upon behavior over the preceding 3 months. The CTS2 violent behaviour outcome was dichotomized, where 0 was defined as no reported violence and 1 was defined as one or more episodes of any type of violence. The Norwegian version of the CTS2 has been used in a Norwegian student population [21].
Sample size
The power calculation was based on an assumption of a Poisson distributed number of violent behaviour events per individual on the CTS2. We expected some within-individual clustering but we did not find any estimates of such clustering or the actual level of violence events in previous studies. Hence, we assumed a within-individual clustering of 30%. With a statistical significance level of 5% when comparing the two groups, we anticipated being able to detect a 20% difference (10 events vs 8 events) with 80% power with a sample size of 134 (67 individuals in each study arm).
Randomization
The participants were allocated to intervention using an Internet-based computer program provided independently by the Research Trial Service Centre at the Norwegian University of Science and Technology (www.webcrf.medisin.ntnu.no). A block randomization procedure (blocks of 10) with no stratification was used, and the participants were informed of their group allocation immediately after the randomization. Those involved in the trial were blinded to the block sizes.
Blinding
Participant and partner data had unique codes and the analyst was blinded to the identity of participants until finalizing the results. The researchers were blinded to the randomization procedure.
Statistical analyses
The descriptive analyses of baseline characteristics were performed with IBM Corp. SPSS, version 23.0 (IBM Corp. Released 2015. IBM SPSS Statistics for Windows, Version 23.0. Armonk, NY: IBM Corp.). The estimation of the effect of treatment on changes in intimate partner violent behaviour over time was conducted according to the intention-to-treat principle. The primary outcome was analyzed with STATA (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC). Due to a skewed distribution of violent events, we dichotomized the outcome into either ‘any violence’ or ‘no violence’. We combined the time points 3 and 6 months’ follow-up, and 9 and 12 months’ follow-up, because of lower response rates during some of the intermediate assessments. Each follow-up wave was added to the model as a dummy variable (3–6 months and 9–12 months and with baseline as reference). In order to investigate differences between the groups during follow-up, we included interaction terms between group allocation and each registration time point. We estimated the proportion of any intimate partner violence according to time and intervention with 95% confidence intervals at each assessment, using a generalized estimating equation (GEE) model with a logit function (Fig. 2). To alleviate underreporting we compared the participant-scores and the partner-scores on each of the CTS2-items in our calculations and used the larger of the two individual item responses, i.e. the highest reported incidence of violence was included in the analyses.
Sensitivity analyses
We investigated possible differences in loss to follow-up between the groups with a linear regression model where the outcome variable was number of responses during follow-up. The concordance between clients and partners was assessed with a logistic regression analysis. Based on the distribution of the measure of psychological violence, we estimated the mean value of psychological violence between the two interventions based on a linear mixed model. We have also presented the results of an analysis of the outcome measures without any dichotomization, based on a linear mixed model. The linear and logistic mixed models use all the information available and is less sensitive for outcome based missing during follow-up compared to traditional methods [23, 24]. We also analyzed the data in a similar way using client only responses (see Additional file 1).