Intervention and outcomes 

We included all studies reporting the results of a psychological or psychosocial intervention. These were defined broadly. Psychological interventions refer to treatments based on a theory of psychological functioning while psychosocial interventions represent less specific interventions designed to improve mental health through general support, advice and encouragement.11 This includes psychoeducational strategies, cognitive–behavioural therapy, interpersonal psychotherapy, non-directive counselling, supportive interactions and tangible assistance, through individual or group sessions.12 We were interested in 10 outcomes: disturbance, mental well-being, quality of life, recovery, violence/risk, satisfaction, seclusion, symptoms, therapeutic relationship and ward environment. The outcomes were based on the rated importance of outcome domains for forensic mental health research13 and the suitability of assessing these outcomes in forensic inpatient settings. There were no limits on the length of follow-up.

Study design

We only included randomised controlled trial studies. Pilot or feasibility studies were included if a randomised controlled trial design was used.

Data collection

The title and abstracts of studies identified through the search strategy were screened for eligibility by one reviewer using the inclusion criteria. The second reviewer independently screened a 20% random selection of the studies. Any disagreements were resolved through discussions between the two reviewers. Full-text articles were obtained for all studies meeting the initial eligibility criteria. All full-text articles were then examined independently by both reviewers to determine their eligibility for inclusion in the review. Reference lists of all relevant articles were also searched. A data extraction sheet was developed to enable assessment and synthesis of the included studies.

Registration details

We registered the protocol of our systematic review on 21 May 2017 on the PROSPERO database available at http://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42017067099.

Search strategy

The focus of the review was to examine psychological and psychological interventions in forensic mental health settings. CINAHL, MEDLINE, PsycINFO, ScienceDirect and Web of Science databases were searched. We searched for peer-reviewed articles, working papers and policy reports, published in English between 1 January 1990 and 31 May 2018. The following search terms were used:

psychiatr* OR mental*

AND

forensic OR secure OR disordered OR offender*

AND

psycholog* OR psychosocial* OR therap*

AND

quality OR wellbeing OR satisfaction OR recovery OR behavio* OR disturb* OR violen* OR seclusion OR abscond* OR symptom* OR environment OR atmosphere

AND

RCT OR random* OR control* OR placebo OR TAU.

Inclusion and exclusion criteria

We included any psychological or psychosocial intervention given as an individual or group treatment in an inpatient forensic setting. We excluded interventions focusing specifically on a specific cohort (ie, sex offenders) as we were interested in examining approaches appropriate for the vast majority of inpatients in secure units.

We restricted our search criteria to forensic inpatients in low, medium and high secure units aged over 18 years. Our exclusion criteria included community settings where patients received treatment outside of the forensic unit or resided outside of the forensic unit when they were not receiving treatment. However, as detailed in the results section, we decided to include one study where a minority of the participants were residing in the community. We also excluded prison settings that are not deemed places of treatment under the Mental Health Act.

Risk of bias summary 

We used the Cochrane Risk of Bias tool14 to evaluate six domains of bias: selection bias (random sequence of generation and allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective outcome reporting) and other bias. The risk of bias for each domain was rated as high (seriously weakens confidence in the results), low (unlikely to seriously alter the results) or unclear. The risk of bias assessment was conducted by the authors separately. There was an average of 1–2 domain ratings per study where there was an initial disagreement. In all cases, the reviewers discussed and agreed the ratings without involving a third party reviewer.

Data synthesis 

Meta-analysis was initially planned but was considered inappropriate because of the heterogeneity of the identified studies due to: the different characteristics of the participant inpatient populations, the different types of approach used by the intervention and control groups and the different outcome measures being used. We therefore present the results in table form together with a narrative synthesis.

Patient and public involvement

Patients and the public were not involved in the design or analysis of this review.

Study setting and characteristics

The trials’ characteristics are shown in table 1. The trials involved 523 participants with a median sample size of 63, ranging from 14 to 112. Five studies included women with a total of 37 participants accounting for 7.1% of the overall sample. All participants were individually randomised except for one study23 where cluster randomisation was used. Eight studies were conducted in the UK, two in the Netherlands, one in Finland and one in Canada. Four studies were conducted in high secure settings and three studies in medium secure settings. The other two studies were conducted in a combination of high, medium and low secure settings including one study where a minority of the participants were living in the community.21 Four of the studies included participants diagnosed with schizophrenia or a psychotic disorder, three studies included participants with a diagnosis of personality disorder and two studies included participants from both diagnostic groups.

Types of intervention

Five broad types of intervention were undertaken (cognitive–behavioural therapy (CBT), dialectical behaviour therapy (DBT), psychoeducation, schema-focused therapy (SFT) and solution-focused brief therapy (SFBT)).

Time points

All studies detailed the baseline assessments with the scores for the intervention and control group comparable at baseline. The studies also reported assessment scores immediately post treatment (except ref 22), at 3-month post-treatment,15 6-month post-treatment21–23 27 and 1-year post-treatment.18 Doyle et al 19 recorded scores at 6, 12 and 24 months and Bernstein et al 16 at 3, 6, 12, 18, 24, 30 and 36 months.

Outcomes

Nine of the 10 outcomes of interest were reported in the studies. Eight studies reported violence/risk outcomes, four reported symptoms outcomes, three reported quality of life outcomes, three studies reported recovery outcome and two studies reported disturbance with one study reporting on therapeutic relationship/engagement, satisfaction, ward environment/atmosphere and seclusion. There were no reported outcomes for well-being.

Two of studies did not report any raw scores. Doyle et al 19 reported the outcomes at the three different follow-up times (6, 12 and 24 months) with these analysed simultaneously in a repeated measures analysis using all available data and recording the estimated treatment effects (group differences) and p values. Bernstein et al 16 used repeated measures analysis of variance to analyse the effect of SFT versus TAU on Historical Clinical Risk Management- 20 (HCR-20) scores over the course of treatment. They did not analyse other outcome variables because of the low statistical power in the sample.

Overall, there were few significant findings with only 7 reported out of 91 statistical comparisons.

Risk of bias of evidence

The majority of domains had a low risk of bias (figure 2). In relation to the potential of performance bias, we determined that participants and staff would be aware of which arm of the trial they have been allocated but any performance bias would be minimal. We, therefore rated these studies as having a low risk. There were difficulties with recruitment and attrition adding to the limitations of the small sample sizes of the studies. Five of the studies reported problems with recruitment with between 23% and 29.4% of patients deemed as eligible refusing to participate. Three studies were rated as high risk of attrition bias due to incomplete outcome data18 19 27 with two studies reporting over 50% of their intervention group not completing the sessions.18 19

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Risk of bias table.

Six studies were able to limit detection bias through ensuring the blinding of the raters of the outcome assessments. One study where blinding was not performed acknowledged the participants may have shown social desirability bias,25 while another used raters who were blind to patients’ treatment condition status double-scored a subset of these assessments with good levels of inter-rater agreement recorded.13

Main findings

This systematic review found a total of nine published RCTs examining psychological and psychosocial interventions in forensic mental health inpatient settings deliverable to any patient residing in a forensic mental health inpatient setting. The studies were heterogeneous resulting in a narrative review of the main findings. There were few statistically significant findings reported; only 7 out of the 91 comparisons analysed, and none of these significant findings revealed a consistent result. This indicates the current evidence base for supporting any psychological or psychosocial intervention is limited. Table 3 gives some indication of areas where particular interventions may have a positive benefit, though with the lack of significant differences recorded, these findings need to be treated with caution. In general, psychoeducational approaches reported improvements in recovery and symptom outcomes and poorer findings for quality of life outcomes. The CBT interventions noted improved findings for absconding and symptoms outcomes, though the impact on violence/risk was more equivocal. A similar finding is noted in relation to the SFT intervention with an equal amount of better and worse outcomes recorded for measures of violence/risk. The DBT intervention show promising results for reducing violence/risk, while the SFBT approach reported improved quality of life, therapeutic relationships and reduced disturbed behaviour. However, the results of both interventions are based on single small-scale studies indicating more extensive studies are required to produce clearer evidence. This review suggests that psychological and psychosocial interventions do not reduce violence/risk in this group of patients, though there is some tentative support that the interventions may improve mental health symptoms.

A number of other factors may have contributed to these findings: individual study designs were quite different, a variety of different secure settings were included with two studies recruiting from different levels of security21 27 and most studies were recruited from multiple sites. The interventions may have had different impacts due to differences in the therapeutic uses of security and related legal governance systems.2 The study sample sizes were small ranging from 14 to 112 participants. This lack of statistical power limited the ability of the study to detect treatment differences.21 The representativeness of the findings was reduced through most studies only including participants with either a diagnosis of psychosis or personality disorder and by the small number of women participating. The paucity of women participants in forensic research has been viewed as indicative of the realities of research undertaken in this area where basic services to women were often poor or lacking.35 One study noted the significant number of women withdrawing from the study when compared with men and suggested examining reasons for higher dropout rates and whether specific support was required during the intervention.23 The time line of the intervention varied considerably from eight weekly sessions of psychoeducation15 to twice weekly sessions of schema therapy for 3 years.16 The recording of the control group intervention varied greatly and consisted of widely differing approaches. There were also differences between the number of treatment sessions offered to the intervention and control groups.  Only one study offered both groups the same number of therapy sessions.21 It is possible that the different treatment intensity may have influenced the outcomes.16 The lack of standardised outcomes was also problematic. Thirty-one outcomes measures were used, with only five measures used more than once, making it difficult to draw conclusions across studies.13

Other reviews of research in forensic mental health settings have reported similar difficulties preventing a homogenous dataset36–39 with few studies with enough similarities to each other to draw firm conclusions regarding the impact of interventions.36 Continuing with small-scale research with mentally disordered offenders (MDOs) is questionable due to these studies being underpowered and unlikely to detect differences.37 Future studies would benefit from larger sample sizes that include representative groups of the forensic inpatient population. It is likely that multisite studies will need to be undertaken with the impact of different environments reviewed as part of the study. To increase the homogeneity of studies, future studies also need similar participants, interventions and outcome measures.36 Using measures that are familiar in practice might be a productive way of developing standardised outcomes.13

Strengths

The majority of studies included in the review had a low risk of bias, indicating it is possible to conduct well-designed RCTs in forensic mental health inpatient settings.10 RCTs remain the gold standard for investigating the effectiveness of treatments and are needed to determine beneficial interventions for this group of patients.16 The randomisation procedures worked well in the majority of the studies. Seven studies reported on the fidelity of the intervention approach with staff trained in the specific procedures. The intervention approaches were competently performed with only one study19 noting that therapists providing the intervention may not have met relevant standards. Most studies were also able to blind researchers to allocation status when assessing outcomes.

Limitations

The review excluded non-English language publications that may have led to some relevant research not being included in the review. Some limitations were noted with recruitment and attrition. Five studies reported that approximately 25% or more of the patients approached declined to participate. It was likely the patients who declined to participate were more unwell and/or antisocial, and these factors might have influenced treatment outcomes.40 Attrition was also a problem, which is not surprising considering the high levels of anti-social behaviour and non-compliance prevalent in this cohort.41 Overall, 25% of participants withdrew or dropped out of the studies. Those limitations that took longer to complete or that required a high level of weekly committment were more likely to record a greater number of dropouts and withdrawals.