Location of studies

All studies were conducted in the USA, with the exception of the study by Jaberghaderi et al.,¹⁰² which was conducted in Iran, and King et al.,¹⁰³ which was undertaken in Australia.

Study size

Five studies^{90,99,101–103} had small samples sizes ranging from 18 to 63 participants. A multisite trial by Cohen 2004,^95,96 had a sample of 229 participants. The remainder ranged from 82 to 210 participants.^{91–94,97,98,100} As a result, the meta-analyses we conducted were not sufficiently powered to detect small, but potentially important, effects. Baseline differences in these studies also proved problematic in drawing any conclusions based on end-point data, as it did for almost all included studies.

Participants

Maltreatment

The range of abuse experienced by participants was broad, and differently reported, but the following picture of participants emerged. Most were abused by men known to them. The majority of perpetrators were family members. In three studies,^89,90,93,94 approximately half of the children and young people had experienced oral, vaginal or anal penetration. In the study^91,92 dealing with the youngest participants, the percentage that had experienced vaginal or anal intercourse was 26%.^91,92 In the study of children aged 2–8 years,⁹⁹ the number reported to have experienced penile penetration was 16%. Participants in all studies ranged from those who had experienced one incident of abuse to those who had experienced multiple incidents, sometimes over many years. Many participants also reported the use of force, or threat of force. Not all studies reported detailed abuse data, for example Deblinger 2001⁹⁹ or Jaberghaderi 2004.¹⁰² See Table 3 for a profile of participants in each study.

Inclusion criteria

All studies had inclusion criteria that specified contact sexual abuse. All but two trials^101,102 made the independent substantiation of sexual abuse an inclusion criterion. Most set cut-off points on the time of last episode of abuse as an inclusion criterion, ranging from 3 months¹⁰¹ through 6 months^91–94 to 2 or 3 years.¹⁰³ Although Deblinger et al.⁹⁹ did not set a time limit, the authors report that the mean age of the children was 5.45 years (SD 1.47 years) and the mean age of first experience of sexual abuse was 4.5 years (SD 1.47 years), based on mothers’ estimates. The report by Berliner and Saunders⁸⁹ did not specify inclusion or exclusion criteria, but all participants were said to have provided statements, substantiated by independent assessment, that they had been sexually abused. The Jaberghaderi et al. study¹⁰² required that girls had experienced sexual abuse at ≥ 6 months prior to the study.

The presence and severity of symptoms as inclusion criteria were highly variable. Six studies^{91,92,95–98,101–103} reported the presence of particular symptomatology thresholds as an inclusion criterion. Cohen et al.^91,92 required a minimal level of symptomatology defined as a Weekly Behavior Report total behaviour score of > 7 or any sexually inappropriate behaviour reported on the Child Sexual Behavior Inventory (CSBI).²⁵⁹ Cohen et al.^95,96 stipulated that participants had to meet five criteria for sexual abuse-related Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV)-defined PTSD, including at least one in each of the three clusters (re-experiencing, avoidance or numbing and hyperarousal). Deblinger et al.^97,98 required the presence of three PTSD symptoms, including at least one symptom of avoidance or re-experiencing the phenomenon. Investigators decided to take both children who met full Diagnostic and Statistical Manual of Mental Disorders-Third Edition, Revised criteria for PTSD and those with partial PTSD symptoms because of the possibility of delayed onset of episodic course. King et al.¹⁰³ required that the children met diagnostic criteria for PTSD or provided evidence of high risk of developing the disorder. Foa et al.¹⁰¹ required a primary diagnosis (DSM-IV, Text Revision) of chronic or subthreshold PSTD. Jaberghaderi et al.¹⁰² recruited girls whose scores on the Child Report of Post-traumatic Symptoms (CROPS)²²⁹ indicated a clinically significant level of post-traumatic symptoms.

Interventions and comparison

Two studies^99,268 provided a group-based intervention, and the remainder provided individual therapy (IT).

Group treatments

Individual treatments

Those studies headed by Cohen and Deblinger^95,96 are essentially evaluations of a manualised programme first developed by the authors in the early nineties.²⁸³

Cohen 1995^91,92 evaluated a manualised, short-term treatment model designed for sexually abused children and their parents, named Cognitive–Behavioural Therapy-Sexually Abused Preschool Children. Children receive safety education and assertiveness training, are helped to identify appropriate compared with inappropriate touching, and to deal with attributions regarding the abuse, ambivalent feelings towards the perpetrator, regressive and inappropriate behaviours, and fear and anxiety. Specific issues for parents include ambivalence in their belief in the child’s account, ambivalent feelings towards the perpetrator, attributions regarding the abuse, concerns that the child is ‘damaged’, how to provide appropriate emotional support for the child and manage inappropriate child behaviours, fear and anxiety. Interventions include the use of cognitive reframing, thought-stopping, positive imagery, contingency reinforcement programmes, parent management training and problem-solving. Psychoeducation and support are embedded in the programme.

Cohen 1998^93,94 evaluated a programme entitled Sexual Abuse-Specific Cognitive–Behavioural Therapy (SAS-CBT). The programme is not described in detail, but it was designed specifically to address depression, anxiety and behavioural difficulties. SAS-CBT addressed feelings of helplessness (including not being believed), distorted attributions (self-blame) about the abuse and other negative events, feeling damaged/different, and consequent low self-esteem. It incorporated anxiety reduction techniques, such as thought replacement, positive imagery/relaxation, enhancement of safety and management of intrusive thoughts. It helped children to address behavioural problems by teaching them about the connections between thoughts, feelings and behaviour, management techniques and problem-solving skills. The focus in the parents’ groups was on reducing their emotional distress (again, including addressing distorted attributions, anxiety and anger), enhancing their ability to support their child and behaviour management.

Deblinger 1996^97,98 explored variations of a programme described as similar to that of Cohen and Mannarino.^91,92 Participants were assigned to one of three experimental conditions: child only, parent only or combined child and parent. Children in all of the experimental arms received an intervention that included GE, modelling, education, coping and body safety skills. GE was described as the cornerstone of the intervention aimed at helping the children to disconnect the associations frequently made between highly negative emotions and abuse-related thoughts, discussion and other reminders. Parents in the experimental arms were taught how to respond therapeutically to their children’s behaviours and needs, that is, how to reduce their fears and avoidance behaviours (through the use of modelling, GE and processing exercises); how to analyse their own interactions with their children behaviourally, thus identifying those situations when they might inadvertently have reinforced problem behaviours and the maintenance of PTSD symptoms; and child management skills.

Cohen 2004^95,96 delivered the same manualised TF-CBT intervention used in earlier studies by this team, but in a more representative sample of children across two sites. This manualised intervention also forms the basis of the study conducted by Deblinger et al.¹⁰⁰ in 2011. In this paper¹⁰⁰ the authors describe the TF-CBT intervention evaluated in this study as including components ‘that spell out the acronym PRACTICE:

• Psychoeducation and parenting
• Relaxation
• Affective modulation
• Cognitive coping
• Trauma narrative (TN)
• In vivo exposure
• Conjoint parent–child sessions, and
• Enhancing safety and future development’ (p. 69; © 2010 Wiley-Liss, Inc. Reproduced with permission).

In this study,¹⁰⁰ the authors were concerned to investigate the importance of the TN to effective treatment of children with PTSD. This four-arm trial compared two versions of TF-CBT (as described for Deblinger 2001⁹⁹), one with, and one without, the inclusion of the TN component and at the same time manipulated the length of treatment and degree of time given to the TN. The authors report that in all conditions both children and parents received psychoeducation about CSA and skill-building (e.g. relaxation, affective modulation, cognitive coping and body safety training), as well as parenting skills training. However, only those children assigned to the two TN groups ‘were actively encouraged to develop a detailed narrative about the sexual abuse and related experiences, which they processed and reviewed with the therapist as well as their non-offending parent’ (p. 69).¹⁰⁰ Children in the eight-session TN condition spent three to four sessions on the TN component; this was at least doubled in the 16-session condition.

King et al.¹⁰³ evaluated the effectiveness of two CBT interventions, both of which the authors say were particularly influenced by the work of Deblinger et al.^208,485 The first was a child-only intervention. This began with a session that specified the problem areas, presented the rationale for the programme and set goals. The following three sessions focused on teaching coping skills to enable children to deal with disturbing memories of abuse and their feelings of anxiety and guilt (relaxation training, behaviour rehearsal and cognitive therapy). Sessions 5 through to 18 focused on graded exposure, and sessions 19–20 on relapse prevention and education, including personal safety skills. The second intervention was family CBT, in which the child received the programme outlined above, and non-offending mothers also received a CBT intervention. The parent intervention began with the rationale of the programme and issues relating to CSA, followed by nine sessions ‘on the development of parent–child communication skills in order to facilitate listening and problem sharing and to overcome avoidance of abuse-related discussion within the family’ (p. 1350).¹⁰³ The remaining 10 sessions focused on child behaviour management, including antecedent stimulus control and contingency management. Parents were encouraged to monitor their own emotional responses in order to provide an appropriate coping model for the child.

Celano et al.⁹⁰ evaluated the impact of the Recovering from Abuse Program, an eight-session group that focused on children’s maladaptive beliefs, affects and behaviour along four dimensions: self-blame/stigmatisation; betrayal; traumatic sexualisation and powerlessness.

Jaberghaderi 2004¹⁰² compared individual CBT with EMDR, and Foa 2013¹⁰¹ compared exposure therapy with supportive counselling.

Foa 2013¹⁰¹ evaluated the effectiveness of prolonged exposure therapy, delivered in eight modules, comprising (1) explaining the treatment rationale; (2) establishing an index of trauma and teaching participants breathing control; (3) presenting common reactions to trauma; (4) explaining the rationale for in vivo exposure, establishing an in vivo hierarchy and arranging homework for the participant; (5) two to five sessions of imaginal exposure lasting between 15 and 45 minutes, combined with reprocessing of the experiences; (6) four to seven further sessions of imaginal exposure centred on the most extreme periods of trauma; (7) generalisation of newly acquired skills and relapse prevention; and (8) a final project, ‘such as making booklets about the trauma and the gains made in treatment’ (Foa 2013¹⁰¹ p. 2652).

Number and duration of sessions

Outcomes assessed using the same measure

When the impact of intervention on parenting practices was assessed, the measure used was the Parenting Practices Questionnaire (PPQ).²⁷⁴

In each study that examined child depression, child sexual behaviour and child behaviour, the same measures were used, namely the CDI,⁸⁰ the CSBI²⁵⁹ and the CBCL,²⁶⁰ respectively, although different authors may cite different sources for the same measure.

Outcomes assessed using different measures

Sequence generation

Sequence allocation was deemed sufficiently robust to be judged low risk of bias in four studies.^{89,91–94,99} The Jabergadheri et al.¹⁰² study was judged as being of ‘unclear’ risk of bias because the authors write ‘Participants were randomly assigned to treatment condition, with some adjustments to promote equivalence between groups’ but then go on to describe a blocked randomisation approach (p. 361). The remaining studies were also judged ‘unclear’, as the authors simply report that participants were randomly assigned.^{90,95–98,100–103}

Allocation concealment

Only two studies^89,101 described steps taken to conceal allocation or stated clearly that allocation had been concealed. The Berliner and Saunders⁸⁹ study reported that ‘Assigned therapists and other staff were blind to the random assignment schedule’ (p. 299) and the Foa et al.¹⁰¹ study reported that ‘On completing the preparatory phase but prior to the patient beginning treatment, a research assistant consulted the randomisation table and notified the therapist of the patient’s treatment condition’ (p. 2651).

No information was provided in the remaining studies. The Jabergadheri et al.¹⁰² study was assessed as ‘unclear’ on this basis.

Blinding of outcome assessors

Four trials^{90,93,94,101,102} were judged as ‘low risk of bias’ for outcome assessment. Foa et al.¹⁰¹ state that ‘assessment was conducted by two psychologists, blind to assignment’ (p. 2651), and Cohen 1998^93,94 say that ‘The evaluator conducting the initial and follow-up assessments was blind to treatment condition or assignment’ (p. 139); Jaberghaderi et al.¹⁰² says that assessment was conducted ‘by two psychologists, blind to assignment’ (p. 362). Celano et al.⁹⁰ reported that ‘standardised measures were administered by a clinician not involved in the child’s treatment. Additionally, a psychiatrist or psychologist blind to treatment condition rated the child’s overall psychosocial functioning based on child and caretaker interviews’.

The remaining studies^{91,92,95,96,99,100,103} were deemed as ‘high risk of bias’, as they relied wholly on self-report or parent-report measures.

Incomplete outcome data

Four studies^{93–96,100,101} were assessed as ‘low risk of bias’ for incomplete outcome data. All four analysed data on an intention-to-treat (ITT) basis, although Cohen 2004^95,96 used SAS multiple imputation for missing data [release 8.2 (2001); SAS Institute Inc., Cary, NC, USA].

Berliner and Saunders⁸⁹ experienced high levels of attrition and reported on around only 50% of those children who completed at least eight sessions and provided data at one of the follow-up assessment points. The statistical checks undertaken by the authors do not attenuate the likelihood of bias. The levels and approach to attrition was also an issue for Cohen 1996,^91,92 King et al.¹⁰³ and Jabergadheri et al.¹⁰² The impact of missing data was less clear in the studies by Celano et al.,⁹⁰ Deblinger et al.^97,98 and Deblinger et al.,⁹⁹ and these were judged to be ‘unclear’.

Selective outcome reporting

Both Foa et al.¹⁰¹ and Cohen 2004⁹⁵ registered their trials. Foa et al.¹⁰¹ reported findings for the primary outcome listed; there were no secondary outcomes identified on the trial registration (ClinicalTrials.gov identifier NCT02148484). In addition to all outcome measures listed in the trial registration, Cohen et al.⁹⁵ also report post hoc on changes in child behaviour, but overall we judged this trial to be ‘low risk of bias’ on this domain (ClinicalTrials.gov identifier NCT00000383).

In the absence of study protocols it is extremely difficult to assess the risk of selective outcome reporting. In general, most other studies appeared to indicate that they were reporting on all predetermined outcomes. In that respect, almost all were assessed as ‘low risk of bias’. However, Cohen 1998^93,94 did not report the results for one measure in their report of initial outcomes, but did report outcomes for all time points at 1-year follow-up. Deblinger 2011¹⁰⁰ report only the results of analysis of covariance (ANCOVA) for complete sets of data. These two studies^93,94,100 were therefore judged to be of ‘high risk of bias’.

Other sources of bias

No other potential sources of bias were identified in these studies. In Jaberghaderi et al.¹⁰² the therapists delivering the CBT and EMDR were both authors on the paper, but, as each was delivering the therapy in both arms of the trial, we judged this to be of ‘low risk of bias’.

Post-traumatic stress disorder

Six studies^{90,93–99,103} examined the impact of CBT on post-traumatic stress using a variety of scales (see Table 5). These studies^{90,93–99,103} yielded an average reduction of 0.44 SDs based on various child PTSD scales (95% CI 4.43 to 1.53; I² = 46%; p-value for heterogeneity 0.10; τ² = 0.05) immediately after treatment, and three of these^93–98 suggested a reduction of 0.38 SDs (95% CI 0.65 to 0.11; I² = 4%; p-value for heterogeneity 0.35; τ² = 0.00) after at least 1 year (Figure 2).

FIGURE 2

Cognitive–behavioural therapy vs. no CBT for PTSD. df, degrees of freedom; IV, instrumental variable.

In our sensitivity analysis we found that over the range of correlations assumed (ρ= 0, 0.25, 0.5, 0.75, 1), results for PTSD are robust to whether follow-up measures or change score measures are used. This was the case for both post-test outcomes and 1-year outcomes.

Depression

Five studies^{90,93–98,103} looked at the impact of CBT on depression in children using the CDI. When combined in a meta-analysis, these five studies yielded an average reduction of 2.83 points on the CDI immediately after intervention (95% CI –4.53 to 1.13; I² = 22%; p-value for heterogeneity 0.27; τ² = 0.84) (Figure 3).

FIGURE 3

Cognitive–behavioural therapy vs. no CBT for depression. df, degrees of freedom; IV, instrumental variable.

Four of these studies^89,93–98 sustained an average decrease of 1.42 points (95% CI –2.91 to 0.06; I² = 0%; p = 0.06; τ² = 0.0) after at least 1 year (see Figure 3). A decrease of 2.9 on the CDI represents something in excess of a modest change in a scale that previous studies^683,684 indicate has a SD of approximately 7. This represents a small to moderate effect size, broadly equivalent to that seen on the anxiety scales in meta-analyses in this review.

In our sensitivity analysis, we found that for both post-test measures and 1-year follow-up measures, the results were closer to ‘no effect’ when using change scores rather than follow-up scores. CIs became wider as the assumed correlation reduced, but even when the correlation is assumed to be 1, the estimated mean difference (MD) is –1.98 for post-test scores with a 95% CI of –4.59 to 0.63, and –0.51 for 1-year follow-up scores with a 95% CI of –3.0 to 2.0.

Anxiety

Five studies^{89,93–98,103} examined the impact of CBT on anxiety. These yielded an average decrease of 0.23 SDs on various child anxiety scales (95% CI 0.03 to 0.42; I² = 0%; p-value for heterogeneity 0.84; τ² = 0.0) immediately after treatment, and four of these^89,93–98 reported a sustained decrease of 0.28 SDs (95% CI –0.52 to –0.04; I² = 0%; p-value for heterogeneity = 0.62; τ² = 0.0) after at least 1 year (Figure 4).

FIGURE 4

Cognitive–behavioural therapy vs. no CBT for anxiety. df, degrees of freedom; IV, instrumental variable.

In our sensitivity analysis, we found that for post-test measures, the results were closer to ‘no effect’ when using change scores rather than follow-up scores. CIs increased as the assumed correlation reduced, but, even when the correlation is assumed to be 1, the estimated SMD is –0.19 with a 95% CI of –0.51 to 0.12. For 1-year follow-up measures, results were robust with similar results using change scores as with follow-up scores, showing no evidence of effect.

Sexualised behaviour

Five studies^{89,91–96,99} provided conflicting evidence on the effectiveness of CBT in the domain of child behaviour problems, assessed using the CSBI (I² = 67%, p-value for heterogeneity 0.02; τ² = 6.81). Two studies^89,99 observed increases of 4.7 and 1.7 points and three studies^91–96 observed decreases, one^91,92 of which was statistically significant. In a meta-analysis, there was no evidence of an effect on average (mean decrease of –0.65 points, 95% CI –3.53 to 2.24 points).

Four studies^89,91–96 provided longer-term data. The first of these observed a much smaller increase than the same study in the short term. Overall, the average effect found at the 3- to 6-month assessment point was similar to that immediately after treatment but was not statistically significant (–0.46 points, 95% CI –5.68 to 4.76 points; Figure 5). Note that Cohen 2004^95,96 did not report data for this outcome because it was not statistically significant. Although we have not been able to retrieve these data, their inclusion is highly unlikely to change the overall conclusion for this outcome.

FIGURE 5

Cognitive–behavioural therapy vs. no CBT for sexualised behaviour. df, degrees of freedom; IV, instrumental variable.

Externalising behaviour (e.g. aggression, ‘acting out’)

Seven studies^89–98,103 provided data on the CBCL, using the externalising behaviour scale. A meta-analysis of standardised differences in means (owing different scoring systems being used for the scale) did not provide evidence of a beneficial effect on average (decrease of 0.12 SDs, 95% CI –0.40 to 0.17). However, the results were inconsistent (I² = 58%; p-value for heterogeneity 0.03; τ² = 0.08), with one study⁸⁹ observing a statistically significant increase and one study observing a statistically significant decrease.^97,98

Only five studies^90–98 provided longer-term data from which no clear picture merged of either benefit or harm (Figure 6).

FIGURE 6

Cognitive–behavioural therapy vs. no CBT for child externalising behaviour. df, degrees of freedom; IV, instrumental variable.

Behaviour management skills of parents

Data from three studies^95–99 provide information on the effects of involving parents in CBT interventions with their children on their ability to manage their children’s behaviour. All three studies used the PPQ to assess change. Only two^95,96,99 of these studies reported follow-up data for 1 year post treatment. A meta-analysis of outcome data from these two studies indicates a decrease in mean scores of –0.89 1 year after treatment (95% CI –4.89 to 3.11). The long-term effects were not statistically significant but raise doubt about the maintenance of change shown in the post-treatment results, which favoured CBT (Figure 7).

FIGURE 7

Cognitive–behavioural therapy vs. no CBT for behaviour management skills of parents. df, degrees of freedom; IV, instrumental variable.

Two studies^90,95,96 used, respectively, the Parental Reaction to Incest Disclosure Scale and the Parental Support Questionnaire to measure parental belief of their children and support for them. A meta-analysis of standardised differences in means gave a statistically significant increase of 0.3 SDs in favour of CBT (95% CI 0.03 to 0.57) (Figure 8).

FIGURE 8

Cognitive–behavioural therapy vs. no CBT for parental support to child. df, degrees of freedom; IV, instrumental variable.

Only one study⁹⁰ examined parental attributions. In this study,⁹⁰ the author reported small, statistically non-significant improvements on four aspects of parental attributions, using the Parental Attribution Scale. Parents who had been involved in the CBT arm of this study⁹⁰ were less likely to blame themselves or their child for what had happened, were slightly more optimistic about their child’s future than those in the TAU group and more likely to hold the perpetrator responsible. However, the CIs were very wide, crossing the line of no effect.

The Parent Emotional Reaction Questionnaire (PERQ) is designed to assess stressful parental emotional reactions to the sexual abuse of their children. Parents are asked to endorse the frequency of specific reactions including fear, sadness, guilt, anger, embarrassment, shame and emotional preoccupation. No psychometric data are currently available for this measure. In the two studies^95,96,99 that used the PERQ we found a decrease of seven points in parents’ negative reactions (95% CI 3.8 to 10.1). Cohen 2004^95,96 measured outcomes longer term, and observed a smaller but still statistically significant, decrease of 4.6 points at 1 year.

Additional results: studies of cognitive–behavioural therapy for sexually abused children

The above meta-analyses incorporate most of the data available from the randomised trials of CBT for children who have been sexually abused. Some studies reported on outcomes that are not covered above, or presented data in ways that could not be incorporated. These are briefly summarised here.

King et al.¹⁰³ assessed the effectiveness of CBT for improving children’s self-efficacy as measured by the (self-report) Coping Questionnaire for Sexually Abused Children (developed by the author) but reported no significant differences between the group receiving CBT and a wait-list control group.

Deblinger 2001⁹⁹ reported a significant difference in favour of CBT for the total score on the CBCL [repeated-measures multivariate analysis of variance (MANOVA), time/time × group)]. They commented that children in neither group were encouraged to talk in detail about their abusive experiences, owing to their young age, which might account for the smaller reductions in PTSD among the CBT group compared with the control group.

Jaberghaderi et al.¹⁰² compared the effectiveness of CBT with EMDR, using two outcome measures. The first was a broad-spectrum, self-report questionnaire-based measure of post-traumatic symptoms (rather than PTSD), CROPS and PROPS.²⁸⁸ The second was a teacher report scale of potential mental disturbance: the Rutter Teacher Scale.²⁸⁹ For post-traumatic symptoms the authors reported large effect sizes (pre- to post-treatment) for both CBT and EMDR and a moderate effect size for the behavioural measure (the Rutter Scale). No significant differences were found between the two treatments. This was a small study¹⁰² (n = 14) with no follow-up and was one of the few studies conducted outside the USA.

Foa et al.¹⁰¹ assessed the impact of prolonged exposure therapy using piecewise linear mixed models (LMMs) for continuous data and generalised LMMs for dichotomous data. The authors reported that those treated with prolonged exposure derived greater benefit than those who received supportive counselling, even when delivered by counsellors who typically delivered that form of therapy. Those who received prolonged exposure demonstrated greater improvements on the PTSD symptom severity scale (primary outcome) and on all secondary outcomes, namely self-reported PTSD severity, depression and global functioning. Treatment differences were maintained at 1-year follow-up.

Summary

We identified 11 studies of CBT interventions for children and young people who have been sexually abused. Six studies^{90–96,99,101} compared CBT with supportive, non-directive therapy and two studies^97,98,103 compared CBT with no-treatment controls (community and wait-list). One study¹⁰² compared CBT with another treatment (EMDR) and two studies^89,90 compared variations of CBT treatments, that is CBT with and without a focus on SIT and a focus on fear and anxiety,⁸⁹ or different exposures of CBT with or without a TN component.¹⁰⁰

For children who have been sexually abused, various adaptations of CBT, some offered individually, some in groups and some including work with parents proved of some benefit in reductions in PTSD, depression and anxiety, which were sustained at 1 year post treatment. One¹⁰² small study reported equal benefit for CBT and EMDR in reducing PTSD symptoms. There was no evidence of benefit of CBT in reducing sexualised and externalising behaviours. Regarding changes in parents, there was evidence of some improvement in parents’ management of children’s behaviour and support for children, and some change in parents’ attributions regarding the abuse. No harms were reported in any study, but no study set out specifically to examine harms.

This evidence is in line with the conclusions of earlier reviews (e.g. Macdonald et al.,⁶⁸⁵ Harvey and Taylor⁶⁸⁶ and de Medeiros Passarela et al.⁶⁸⁷), suggesting that these approaches may be beneficial compared with non-directive, supportive therapies, but the evidence base remains limited.

Completeness and applicability

None of these studies was conducted in the UK, with most undertaken within the USA. However, the profile of participants is clinically comparable to the population of children who might benefit from child and mental health services in the UK, and the therapies evaluated are recognised, and available, in the UK.

Quality of the evidence

Studies of CBT for children who have been sexually abused could be improved, and would benefit from careful and explicit reporting against Consolidated Standards of Reporting Trials (CONSORT) guidelines. Of some concern is that the field is somewhat dominated by a small team of US researchers who evaluate a version of CBT that they themselves developed.

Location

All three^106–109 studies took place in the USA.

Sample sizes

Samples sizes at recruitment were 75,¹⁰⁹ 55^107,108 and 12.¹⁰⁶

Participants

Maltreatment

Participants in all three^106–109 studies had experienced physical abuse or (four cases in the Kolko^107,108) study severe or frequent forms of physical discipline with a risk of injury.

Interventions and comparison

Kolko^107,108 evaluated the impact of two interventions: a manualised FT treatment for physically abused children and a manualised CBT intervention for individual children and their parent(s). In the CBT arm, children and parents received therapy from separate therapists, who implemented parallel protocols based on social learning principles that were designed to address cognitive, affective and behavioural problems. Treatment for the children addressed their perceptions of family stress and their environments; training to develop coping and self-control skills; and interpersonal effectiveness. The intervention for parents focused on their views on violence and physical punishment, their attribution style and expectations, self-control and contingency management. There was throughout an emphasis on teaching intrapersonal and interpersonal skills. The FT was based on Belsky’s²²⁷ interactional or ecological model approach to child maltreatment. It sought to ‘enhance the cooperation and motivation of family members by promoting understanding of coercive behaviour and by teaching the family positive communication skills and how to solve problems together’ (p. 326).²²⁷

The control group received routine community services from providers not associated with the project, as mandated by family service workers. Services were based on an extensive risk assessment and included ‘home visits to provide support and information, family skills specialists who taught homemaking and related skills, and parenting information and support groups’¹⁰⁸ plus regular telephone contact from the caseworker (p. 326).

Runyon et al.¹⁰⁹ assessed the added value of providing treatment to children [combined parent–child cognitive–behavioural therapy (CPC-CBT)], as well as parents (parent-only CBT). Children in the CPC-CBT arm received an intervention covering psychoeducation; affect regulation; coping skills; cognitive coping; assertiveness skills and anger management; general safety skills; application of skills; development of a personal safety plan; role perspective-taking skills; problem-solving skills; preparing a letter of praise; developing a TN; and agreeing and sharing a joint TN (with parents). In addition to the usual parents’ programme (see Comparisons), parents in the experimental arm received input on parent training with the child; refinement and rehearsal of personal safety plans with the child; attention to abuse clarification and the development of the joint TN plus coaching in parent–child interactions; behaviour rehearsal of coping skills; parenting skills and safety plan; sharing of TN; and abuse clarification.

The parent-only CBT control arm received an intervention that comprised disclosure of the referral incident, engagement and assessing parents’ goals, motivational interviewing and commitment to no violence (two sessions), followed by psychoeducation and an introduction to anger management skills (two sessions), continuation of coping skills (three sessions), review and applications of skills, including ABCs of parent–child interactions, development of personal safety plans (three sessions); review of ABCs of parent–child interactions and integration/generalisation of skills (four sessions) and parent training (parent only, two sessions).

LeSure-Lester¹⁰⁶ provided CBT to groups of two to three participants. The intervention was designed to teach participants the feelings associated with anger and aggression, relaxation and self-talk, and alternative ways of coping. It comprised a three-stage education and training module, beginning with education relaxation techniques. This was followed by education and GE and then education and anger control. The therapist used vignettes (as the basis for discussions) that reflected the participants’ natural environments, that is, the residential setting. Participants in the control arm received ‘traditional indirect therapy’ consisting of open-ended discussions and communications of the participants’ self-reports of activities and current events in their daily lives. The therapists were said to have exercised warmth, empathy and genuineness – no specific coping skills were taught.

Post-traumatic stress disorder

Only Runyon et al.¹⁰⁹ assessed the impact of intervention on PTSD, using the KSADS-PL²⁶⁸ (see Post-traumatic stress disorder). This was the primary outcome in this study.¹⁰⁹

Depression

Kolko^107,108 used the CDI⁸⁰ to assess the intervention’s impact on depression.

Behaviour

Runyon et al.¹⁰⁹ and Kolko^107,108 both used the CBCL²⁶⁰ to assess change in child behaviour, along with the Child Conflict Index (CCI).⁶⁸⁸

LeSure-Lester¹⁰⁶ used a rating system of behavioural appropriateness developed within the residential setting, and comprising aggression towards peers, towards staff and compliance with house rules (all scored low or high).

Sequence generation

Kolko^107,108 used a computer-generated procedure based on Efron’s biased coin toss and was assessed as low risk of bias. Runyon et al.¹⁰⁹ used a computer program to randomly determine the treatment type for each group, and on that basis was deemed low risk of bias. No information was provided by LeSure-Lester¹⁰⁶ and so this study was judged unclear.

Allocation concealment

The LeSure-Lester study¹⁰⁶ was judged as ‘high risk’, as the author was both the provider of the intervention and the researcher. The project co-ordinator in the Runyon et al.¹⁰⁹ study was blind at pre-treatment so we concluded that allocation was concealed at this point (low risk of bias). Kolko^107,108 provided no information and so was judged unclear risk of bias.

Blinding of participants and personnel

High risk: participants and personnel were not blinded in any of the three trials.^106–109

Blinding of outcome assessors

Runyon et al.¹⁰⁹ stated that the project co-ordinator (who conducted all assessments) ‘remained blind, to the extent possible, to condition assignment’ but it is not clear to what extent this was achieved. The determination of unclear risk of bias was made. Risk of bias was high in the study by LeSure-Lester¹⁰⁶ as outcome data were provided by staff in the residential care home. Kolko¹⁰⁸ reported that ‘[P]roject staff administering the assessment protocol were unaware of the treatment conditions to which participants were assigned’ and was judged low risk of bias, although the same data were self-reported.

Incomplete outcome data

Attrition in the Kolko^107,108 study was around 6% (one case) in the FT arm, 20% (five cases) in the CBT arm and 17% (two cases) in the routine community services groups. Although these cases did not appear to be different in characteristics, the differential attrition between the two active treatment arms resulted in a judgement of high risk of bias (the reasons for dropout might be related to the intervention). No attrition occurred in the LeSure-Lester¹⁰⁶ study, so this was judged as ‘low risk’. The Runyon et al.¹⁰⁹ study was judged as ‘high risk of bias’: 25% children who completed 3 weeks’ therapy did not provide data post test, and their pre-test scores were carried forward. Attrition increased at follow-up and the authors present no means or SDs at this assessment point.

Selective outcome reporting

The LeSure-Lester¹⁰⁶ study was assessed as low risk for selective outcome reporting. This study¹⁰⁶ is small, modest in design and implementation, and data are given on those outcomes that the intervention was seeking to address (using ‘home-made’ measures). Kolko^107,108 and Runyon et al.¹⁰⁹ appear to have reported on all outcomes covered in the methods section, but, in the absence of a published protocol for either study¹⁰⁶ both were assessed as unclear for this domain.

Other sources of bias

LeSure-Lester’s¹⁰⁶ study was compromised by the conflation of roles held by the author, that is, researcher and therapist.

Post-traumatic stress disorder

The Runyon et al.¹⁰⁹ study reported significant pre- to post-improvement on the total number of PTSD symptoms among all children, with the adjusted mean post-test scores for those in the combined CBT parent and CBT child group being significantly lower than those for the parent-only CBT group.

Child depression

In the Kolko^107,108 study, children’s reports on the CDI⁸⁰ indicated a significant reduction in severity of depressive symptoms over time [χ² = 16.01(3); p < 0.001], but reports were said to be ‘generally low and similar across time, indicating no significant group differences’¹⁰⁸ (p. 333).

Child behaviour

As measured by the Youth Self-Report (YSR) of the CBCL, children in all three groups in the Kolko^107,108 study (CBT, FT and routine services) reported a significant reduction over time in both internalising symptoms [χ² = 33.54(3); p < 0.0001] and externalising symptoms [χ² = 12.26 (3); p < 0.002], with both CBT and FT showing most change on these measures. No effect was found for social competence.

Parent report on the CBCL indicated lower ratings of serious internalising behaviours over time (p < 0.07), particularly for the two treatment arms. Parents reported a significant reduction in externalising behaviour over time [χ² = 9.53(3); p < 0.02]. Based on an inspection of the means over time, CBT appeared to show the greatest initial change and FT the greatest change at follow-up (1 year) compared with routine community services, which showed minimal change during that period.

A significant interaction was reported on the CCI [χ² = 13.12(3); p < 0.04], reflecting the greatest decrease in scores for CBT. This measure (scored by telephone interview with the parent) estimates the presence or absence of common individual behavioural or emotional problems commonly displayed in boys or girls within the previous 24 hours.

Runyon et al.¹⁰⁹ reported significant pre- to post-improvement in internalising and externalising scores (CBCL) for the CBT parent-only condition.

LeSure-Lester¹⁰⁶ reported greater rates of behaviour change from pre-test to post-test for the six adolescents who received CBT. Using the rating system used by staff within the home, these six adolescents demonstrated greater rates of behaviour compliance (t = –5.64; p < 0.001) and less aggression towards staff (t = –4.56; p < 0.001) and other residents (t = –5.64; p < 0.001).

Child global functioning

The Kolko^107,108 study reported a significant increase in KSADS scores over time for all children in the study, with no group differences. This study^107,108 found no difference between the CBT, FT and control groups in reduction over time on fears related to abuse.

Family functioning

Kolko^107,108 reported the results of subscales for the Family Environment Scale (FES) and the Family Assessment Device (FAD). Overall, children and parents in the CBT and FT arms reported more improvement over time than those in routine services.

Summary

We identified only three studies^106–109 of CBT interventions for children and young people who have been physically abused. Each focused on children of somewhat different ages, from middle childhood to adolescence. One study^107,108 compared a CBT intervention for children and their parents with systemic FT; one study¹⁰⁹ compared a CBT intervention for parents with one that included a parallel intervention for children, and the third study¹⁰⁶ compared a small group version of a CBT provided to African American adolescents living in a group home as a direct result of their maltreatment, compared with non-directive group discussions.

Although very different, the three CBT interventions^106–109 shared some common characteristics, namely a focus on children’s thoughts, feelings and behaviour. There was a marked psychoeducational component in both the Runyon et al.¹⁰⁹ and the LeSure-Lester¹⁰⁶ study, aimed at helping children to recognise and understand the consequences of abuse, and develop appropriate coping and problem-solving skills, including the development of skills to minimise risk of abuse^107,108 or personal safety plans.¹⁰⁹

The three^106–109 studies are all extremely small and the overall quality is, at best, moderate in relation to risk of bias. Together with the fact that we can summarise the evidence only narratively, considerable caution is required in interpreting the data. All three^106–109 studies report improvement in children’s internalising and externalising behaviour problems (common sequelae of physical abuse), but one¹⁰⁹ of the studies found an improvement in externalising behaviour in the parent treatment group only.

The one¹⁰⁹ study examining PTSD reported a reduction in symptoms in all children, with the most significant reduction occurring for those where both parents and children received CBT. Depression, examined in one¹⁰⁹ study, reduced over time in both the experimental and comparison groups.

Both CBT and FT generally outperformed routine community services, resulting in greater reductions in children’s externalising behaviour and on child-reported parent-to-child violence and parent-reported child-to-parent violence.

Completeness and applicability

All three^106–109 studies were conducted in the USA. Two^107–109 studies were concerned with families who had come to the attention of services because of maltreatment, but where the child remained in the home. They aimed to minimise the risk of further physical abuse and to address the adverse consequences of past abuse and current maladaptive parenting. In Kolko^107,108 maltreatment was judged as ‘mild/moderate’ in 87% cases, although almost half the families had children that had been hit with an object, and 50% had children who had been smacked with an open hand. In the Runyon et al.¹⁰⁹ study, 53% families had a substantiated allegation of physical abuse or had acknowledged the use of excessive physical punishment (e.g. 65% had hit their children with an object). Many of the children in these three studies would be subject to a child protection plan in the UK, and in the Runyon et al.¹⁰⁹ study (although not in the Kolko^107,108 study) the inclusion criteria required children to have either four PTSD symptoms or an elevated score (T score of ≥ 65) on at least one externalising behaviour subscale on the CBCL.²⁶⁰ In this study, siblings were included in the treatment as long as they too could meet these criteria and were a focus of child protection workers’ concerns. These studies^106–109 therefore have relevance to the UK context, although they are limited in their scope and the evidence base is sparse.

LeSure-Lester¹⁰⁶ evaluated an intervention that was designed specifically to address the aggressive behaviours of boys who had been removed from the family home as a result of maltreatment. Although the focus on addressing maltreatment-related aggression is highly relevant to the UK context, the study¹⁰⁶ says little about the intervention, and the outcome measures focus on resident–staff interaction (with an emphasis on compliance) and peer–peer violence within the home. Although the intervention is reported to have made a significant impact, the size of the study,¹⁰⁶ plus the absence of measures or time periods to indicate the generalisability or likely maintenance of reported behaviour change, mean that its applicability is limited.

Quality of the evidence

Studies of CBT for children who have been physically abused are few in number, poorly reported, and overall of limited quality, although poor reporting may account for many of the identified risks of bias.

Location

All but four^{111,112,116,120} studies were conducted in the USA. The Rushton et al.¹¹⁶ study was conducted in the UK, the Farkas et al.¹²⁰ study in Quebec, Canada, the Church et al.¹¹¹ study in Peru and the Jensen et al.^112,113 study in Norway.

Sample size

Five^{110,111,117,120,121} studies recruited and randomised individual participants who had been maltreated. The Church et al.¹¹¹ study recruited just 16 participants, whereas the studies by Farkas et al.¹²⁰ and Scheck et al.¹²¹ recruited, respectively, 40 and 60 participants to their studies of EMDR and the Shirk et al.¹¹⁷ study randomised 43 adolescents. Champion and Collins¹¹⁰ randomised 559 adolescent women.

The Jensen et al.^112,113 study randomised 156 parents. (Only 135 parents participated in the study.)

The remaining three studies recruited participant pairs. Rushton et al.¹¹⁶ recruited 38 adoptive families. One study by Linares et al.¹¹⁵ enrolled 94 children, with the intervention targeted at foster parent/biological parent pairs, whereas the other Linares et al.¹¹⁴ study recruited 63 biological/foster parent pairs.

Participants

Maltreatment type

In Linares 2006,¹¹⁴ children had experienced physical abuse or neglect, but (by chance) only neglected children were allocated to the control condition, compared with 71% in the intervention group.

In the Linares 2012,¹¹⁵ children had officially substantiated histories of child maltreatment: 77% were neglected and 23% were abused either physically (18%) or sexually (5%). Some children experienced more than one form of maltreatment.

Children in the Church et al.¹¹¹ study had a history of physical, psychological or sexual abuse or neglect/parental abandonment. The majority of participants in the Champion and Collins¹¹⁰ study (76%) had histories of sexual, physical and emotional abuse. This study¹¹⁰ recruited women with abuse histories or histories of STIs (because of the over-representation of maltreatment in the histories of adolescents) and was designed to ‘provide a study sample of adolescents with both a history of STI and abuse’¹¹⁰ (p. 142).

Participants in both EMDR studies^120,121 had histories of maltreatment. Most of those in the Farkas et al.¹²⁰ study had been referred to Youth Protective Services for a variety of forms of parental neglect or abuse, although some were referred for reasons of serious behaviour problems. Most participants had been referred for, or had experienced more than one form of, maltreatment; it was not possible to identify the proportion of participants who had not been maltreated. A total of 90% of participants in the Scheck et al.¹²¹ study reported being victims of physical or emotional abuse as a child, and over half of the traumas reported related to traumatic sexual experiences, such as rape or child molestation.

Adolescents in Jensen et al.^112,113 had been exposed to a range of traumas, including physical and sexual abuse, and witnessing violence.

Interventions and comparisons

The interventions in the studies by Rushton et al.¹¹⁶ and Linares et al.^114,115 were modified versions of Webster-Stratton’s Incredible Years Program (IY).²¹²

Linares et al.¹¹⁴ used the manualised, group-based Parents and Children Basic Series Program (IY, Webster-Stratton et al.²¹⁰) plus a coparenting intervention delivered on individually to biological and foster parent pair and target child, and which focused on learning about each other, practising open communication and negotiating interparental conflict. Therapists used family systems strategies, such as joining, didactic lesson, re-enactment and restructuring.

Linares 2012¹¹⁵ used a subset of the 18 IY manualised lessons contained in the Dina Program for Young Children. Modules were Understanding and Detecting Feelings; Detective Wally Teaches Problem-Solving Steps; and Tiny Turtle Teaches Anger Management, plus a lesson developed for the project and designed to promote a sense of belonging to this foster home – My Homes, My Families.

In both^114,115 of these studies, foster carers in the control group received ‘usual services’.

Rushton et al.¹¹⁶ used the IY programme as a basis for a cognitive–behavioural programme tailored to the needs of adoptive parents, placing an emphasis on the need to conduct daily play sessions with the child and to help adopters when their child rejects their praise or their rewards. First and last sessions were focused, respectively, on getting to know the parents and introducing the programme, and reviewing progress and ending. Other sessions focused on using positive attention to change behaviour; the value of play for establishing positive relationships; using verbal praise; rewards; learning clear commands and boundaries; using ‘ignoring’ to reduce inappropriate behaviour; defining for the child the consequences of undesirable behaviour; ‘time out’ and problem-solving. Adoptive parents in the control group received an educational approach designed by an adoption adviser ‘to improve adopters’ understanding of the meaning of the children’s current behaviour and help them see how past and present might be connected’¹¹⁶ (p. 532), thereby helping adopters to respond more appropriately to challenges.

Church et al.¹¹¹ provided a brief, single-session exposure therapy entitled emotional freedom techniques (EFT), comprising certain components of CBT and exposure therapy combined with a somatic component, having therapists or participants tap their fingers on prescribed acupuncture points. Those in the control group received no treatment.

Champion and Collins¹¹⁰ provided a theory-based [AIDS Risk Reduction Model (ARRM)²⁰⁹] CBT intervention designed to reduce risk-taking behaviour – Project Image (PI). PI is described as ‘grounded in knowledge of the target population’s behaviour and culture . . . Emphasis is placed upon understanding and dealing with male-female power relationships in African-and Mexican American culture’¹¹⁰ (p. 144). The intervention began with a physical examination (for STIs, etc.) followed by an enhanced counselling session (addressing adherence to medication, other treatments, sexual activity, etc.). Intervention participants were then offered two workshop sessions, 1 week apart, followed by group work and further individual counselling. The workshops and group work described have a strong psychoeducational component and a tailored skills component. Control group participants received the physical examination, abuse and enhanced clinical counselling at baseline, plus a follow-up physical examination at the end of the intervention.

Shirk et al.¹¹⁷ evaluated a modified CBT intervention (m-CBT) that combined CBT elements (mood monitoring, cognitive restructuring, relaxation training, activity scheduling and interpersonal problem-solving), with mindfulness-based strategies, such as taking a non-judgemental stance of observing, describing and tolerating trauma-related emotions and cognitions (Linehan et al.⁶⁸⁹). The effectiveness of m-CBT was assessed in relation to usual care (UC), in which therapists agreed to use, with control group participants, the treatment strategies and procedures that they regularly used and believed to be effective in their clinical practice).

The EMDR intervention in the Scheck et al.¹²¹ study consisted of two treatment sessions of 1 hour, 1 week apart. EMDR followed the standard protocol devised by Shapiro.⁶⁹⁰ In this study,¹²¹ EMDR was compared with an active listening intervention.

In Farkas et al.¹²⁰ study, EMDR was combined with motivation–adaptive skills–trauma resolution (MASTR), aimed at addressing conduct problems (Greenwald⁶⁹¹), motivational interviewing and a range of cognitive–behavioural training and coping skills development. MASTR is a trauma-focused treatment package that was developed for use with adolescents with conduct problems, which ‘addresses treatment obstacles by establishing sense of safety within therapy, encouraging clients to be the agents of their change, improving motivation and guiding them towards progressive successes to their goals’¹²⁰ (p. 128). Participants received 12 weekly sessions of 1.5 hours of MASTR/EMDR therapy. They also continued with other forms of individual (14%), family (14%) and group therapy (29%). In this study of EMDR, Farkas et al.¹²⁰ used a ‘routine care’ control group in which participants were exposed to a variety of alternative therapies.

Jensen et al.¹¹² described the TF-CBT programme that they use as a ‘trauma specific treatment consisting of psychoeducation, learning relaxation skills, affective modulation skills, cognitive coping skills, working through the TN, cognitive processing, in vivo mastery of trauma reminders, and enhancing safety and future developments, coupled with the parental component’¹¹² (p. 6). The parental component looked to improve parenting skills and was also used to demonstrate for the parent each treatment component that was provided to the child. Those in the control group received ‘the treatment they (TAU Therapists) considered most suitable in each individual case’¹¹² (p. 6). Almost half of the TAU therapists described their theoretical orientation as psychodynamic, 30% as cognitive behavioural, and around 25% as family/systemic (percentages rounded up). In 35 of the 52 completed TAU cases, parents were involved in some way in more than three sessions of the child’s therapy.

Comparisons

Post-traumatic stress disorder

Jensen et al.^112,113 used two measures of PTSD. The first was the CPSS,²⁸⁶ a self-report questionnaire developed for children aged 10–18 years, which examines post-traumatic stress symptomatology described in the DSM-IV (criterion B, re-experience; criterion C, avoidance; and criterion D, hyperarousal).⁶⁹² The second was the Clinician-Administered PTSD Scale for Children and Adolescents, a structured clinical interview that assesses the frequency and intensity of the 17 DSM-IV-defined PTSD symptoms.^304,305

Scheck et al.¹²¹ and Church et al.¹¹¹ assessed the impact of intervention using the Impact of Events Scale (IES²⁷⁹). Both used the total score; Church et al.¹¹¹ also report outcomes for the memories and avoidance subscales.

Farkas et al.¹²⁰ used two measures of PTSD. First, the relevant module of the Diagnostic Interview Schedule for Children (DISC⁶⁹³) and, second, the TSCC,^325,328 to assess trauma-related difficulties.

Scheck et al.¹²¹ also used the Penn Inventory for Posttraumatic Stress Disorder (PENN³³⁴), a self-report scale that measures symptom severity.

Depression

Jensen et al.^112,113 used the Mood and Feelings Questionnaire⁶⁹⁴ to assess depressive symptoms, as this measures the full range of DSM-IV diagnostic criteria for depressive disorders, and includes items ‘reflecting common affective, cognitive, somatic features of childhood depression’ (p. 361).

Shirk et al.¹¹⁷ and Scheck et al.¹²¹ used, respectively, the Beck Depression Inventory-Second Edition (BDI-II³²⁰) and the BDI³³² to assess the impact of EMDR on depression.

Anxiety

Scheck et al.¹²¹ used the state subscale of the State-Trait Anxiety Inventory (STAI³³³) to measure the impact of EMDR on anxiety.

Jensen et al.^112,113 used the Screen for Child Anxiety Related Disorders (SCARED)³⁰⁷ to measure anxiety symptoms. SCARED is a self-report questionnaire with 41 items covering five specific anxiety disorders: panic disorder or significant somatic symptoms, generalised anxiety disorder, separation anxiety disorder, social anxiety disorder and school avoidance.

Behaviour

Rushton et al.¹¹⁶ and Jensen et al.^112,113 used the SDQ.³⁰⁸ Jensen et al.^112,113 also used visual analogue scales to assess how far an individual child progressed on emotional distress, misbehaviour and attachment. Rushton et al.¹¹⁶ relied on adopter report, whereas Jensen et al.^112,113 used YSR.

In Linares 2012,¹¹⁵ foster parents completed a six-item measure compiled from the CBCL 5-18 aggression subscale,²⁹⁴ and classroom teachers completed a seven-item measure compiled from the 38-item Sutter–Eyberg Student Behaviour Inventory-Revised (SESBI-R³¹¹).

The intervention evaluated in Linares 2006¹¹⁴ was designed to reduce externalising behaviour, and its effectiveness was assessed using three measures, and drawing on foster parent-report and biological parent-report; the CBCL;^294,309 the Eyberg Child Behavior Inventory-Revised³¹⁰ and the SESBI-R.³¹¹

Farkas et al.¹²⁰ used the parent version of the CBCL,^269,294 alongside modules of the DISC to measure conduct disorder (CD) and oppositional defiant disorder.

Risky behaviour

In line with the aim of the intervention, Champion and Collins¹¹⁰ assessed new incidents of STI as a dichotomous variable (yes, no) at off-site, problem or scheduled follow-up visit at 6 and 21 months.

Self-control

Linares 2012¹¹⁵ used a 51-item measure of self-control, developed for this study and administered to foster parent and teacher using parallel versions.

Self-esteem

Scheck et al.¹²¹ examined the impact of EMDR on adolescents’ self-concept, using the Tennessee Self-Concept Scale.³³⁵

Parent–child relationships

Rushton et al.¹¹⁶ used the Expressions of Feeling Questionnaire³¹⁵ to capture the nature and progress of the child’s relationship with the new carers.

Sequence generation

We judged three^112,116,121 studies to be at low risk of bias. In the Rushton et al.,¹¹⁶ study adoptive parents were randomised independently by the clinical trials unit using permuted block randomisation. Jensen et al.^112,113 state that a computer-generated randomised block procedure was used, and Scheck et al.¹²¹ used envelopes filled with papers labelled either EMDR or active listening (AL). These were then shuffled before being numbered 1 through 100. Envelopes were opened (consecutively) during interviews with participants, which took place after the collection of baseline data, thereby identifying to which therapy the participant was allocated.

Linares 2012¹¹⁵ state that children were consecutively identified, assessed and randomly assigned within agencies, but no further information was provided on sequence generation or allocation concealment.

The studies by Champion and Collins,¹¹⁰ Church et al.,¹¹¹ Linares 2006,¹¹⁴ Farkas et al.¹²⁰ and Shirk et al.¹¹⁷ provide no information on sequence generation and were judged to be of unclear risk of bias.

Allocation concealment

None of the RCTs included provided adequate information on allocation concealment, although Rushton et al.¹¹⁶ used a clinical trials unit to randomise participants, so all were judged as being of unclear risk of bias. The remaining eight^{110–112,114,115,117,120,121} studies provide no information on allocation concealment and were therefore judged unclear risk of bias.

Blinding of participants and personnel

We judged all studies^{110–112,115–117,120,121} as being of high risk of bias because no participant or personnel were blinded.

Blinding of outcome assessors

Rushton et al.¹¹⁶ make clear that blinding at follow-up interviews was not possible because involvement in the treatment was the focus of questions. It was therefore assessed as high risk.

Five^{110–112,114,115} studies were assessed as low risk. In both studies by Linares et al.,^114,115 the authors state that intervention and assessment teams were assembled to keep interviewers blind to group assignment. Church et al.¹¹¹ state that data were scored off-site and blind to the statistician. Champion and Collins¹¹⁰ state that group status was revealed only at the end of follow-up interviews.

Jensen et al.^112,113 state that the assessments were computer assisted and conducted by an independent clinician who was blinded to the treatment conditions.

Shirk et al.¹¹⁷ state that post-treatment assessments were made by an independent evaluator, but the depression measure used (BDI) relies on self-completion, and so the study was assessed as being of unclear risk of bias.

In both of the EMDR studies^120,121 the authors state that assessors were blind, but the measures used were largely self- and parent-report, so we judged this as being of high risk of bias.

Incomplete outcome data

There were no missing data in the studies by Church et al.,¹¹¹ Rushton et al.¹¹⁶ or Linares 2012,¹¹⁵ which were therefore judged to be of low risk of bias. Linares 2006¹¹⁴ suffered attrition but reasons for attrition were largely the same (moved, discharged), although more parents in the intervention group refused to provide data post treatment and at follow-up than in the usual services group (eight vs. one). The authors also analysed the data on ITT principles and, overall, we judged the study¹¹⁴ to be of low risk of bias. Attrition in the Jensen et al.^112,113 study was similar across the two arms, and there were no significant differences between the retention group and the attrition group, other than that the attrition group was significantly older than the retention group and the attrition group reported being exposed to significantly higher numbers of different traumatic events. The authors explore reasons for attrition and they are methodical in their analyses, undertaking both ITT analyses and per-protocol analyses, and completed case analyses (defined as those participants who completed at least six sessions). Overall we judged the Jensen et al.^112,113 study to be of low risk of bias.

In the Champion et al.¹¹⁰ study, data are presented for only for 318 out of 409 women at 6-month follow-up (78% unadjusted) and 333 women at 12-month follow-up (81% unadjusted). Given the intervention and the participants, this level of attrition is impressively low, but it was deemed sufficiently large to warrant a judgement of high risk of bias.

Farkas et al.¹²⁰ was also judged high risk of bias, primarily on the grounds of significant attrition. Of 65 adolescents randomised in this study, 15 dropped out by post treatment and a further eight by the 3 months’ follow-up. More young people dropped out of the experimental group than control group during treatment (10/33 vs. 5/32); of these, 2 of the 10 dropped out for reasons related to the treatment (refused to discuss their traumas) and two because they ceased to be in the custody of YPS and their families stopped their participation. All but one of the remaining participants dropped out because they ‘changed their minds’.

In the Scheck et al.¹²¹ study there was considerable attrition post treatment that was not accounted for in the results. We judged this to be high.

Data for 7 out of 43 randomised participants were missing at follow-up in the Shirk et al.¹¹⁷ study [four in m-CBT and three in UC]. Investigation led the authors to conclude that no systematic bias had occurred in attrition, and they conducted their analyses on ITT principles. We therefore assessed this study as low risk of bias.

Selective outcome reporting

Three^110,112,116 studies were registered with ClinicalTrials.gov: Champion and Collins¹¹⁰ (NCT01387646); Rushton et al.¹¹⁶ (NCT04448012) and Jensen et al.^112,113 (NCT00635752).

Champion and Collins¹¹⁰ report findings for the primary outcome (STIs at 12 months) but, in the paper identified for this review, have not yet reported on secondary outcomes (substance use, experience of abuse and frequency of unintended pregnancies). Overall, we judged this as ‘unclear’ risk of bias. The studies by Rushton et al.¹¹⁶ and Jensen et al.^112,113 report on all primary and secondary outcomes and were therefore judged as ‘low’ risk of bias.

In general, the remaining studies appeared to indicate that they were reporting on all predetermined outcomes. However, in the absence of study protocols it is difficult to assess the risk of selective outcome reporting. Therefore, all studies were assessed as ‘unclear’ risk of bias.

Post-traumatic stress disorder

Jensen et al.^112,113 report a mean effect (ITT analyses) of treatment condition on child PTSD (measured by the CPSS) at time 3 (T3), some 7.5 months after treatment began: children in the TF-CBT group scored significantly lower at T3 [mean (M) = 11.34, SD 10.52] than participants in the comparison group [M = 16.87, SD 11.49; d = 0.51, t(154) = 3.30; p = 0.001; with Holm adjustment p = 0.006]. The authors also report a significant time by group interaction effect (F(2) = 5.01; p = 0.007; with Holm adjustment p = 0.037). Both groups showed reductions in PTSD from pre- to post-therapy assessments, and a main effect of treatment condition on functional impairment such that trauma influenced daily functions significantly less (indicated by higher scores) in the TF-CBT group (M = 10.33, SD 1.99) than in the TAU group (M = 9.22, SD 2.09) at the end of therapy [d = –0.55, t(154) = –3.32; p = 0.001; with Holm adjustment p = 0.006]. They also report a main effect of treatment on time in both groups. Analyses of completed cases yielded similar results.

Statistically significant between-group differences were found by Church et al.¹¹¹ in favour of the intervention (EFT). One month after pre-test, participants who had received the intervention demonstrated a statistically significant decrease on both the total score for the IES²⁷⁹ and the two subscales (memories and avoidance). All participants had scored in the clinical range at baseline, and control participants remained in the ‘moderate clinical’ range post test, in contrast to those in the intervention group, none of whom was in the clinical range post test.

Analyses of PENN post-test scores in the Scheck et al.¹²¹ study indicated a significant difference in favour of the EMDR group [F(1,55) = 6.03; p = 0.02]. A similar result was found for the IES [F(1,57) = 9.93; p = 0.002].

Post treatment, Farkas et al.¹²⁰ reported significant improvements in the experimental group (MASTR/EMDR) compared with control group participants for PTSD symptoms as measured by DISC [F(1,40) = 6.05; p = 0.05]. Significant improvements were also reported for the TSCC³²⁵ (trauma-related difficulties) on each of six subscales: stress, anger, depression, dissociation, anxiety and sexual concerns.¹²⁰

Depression

Jensen et al.^112,113 found a main effect of treatment condition on children’s depressive symptoms, with participants in the TF-CBT group (M = 14.40, SD 13.67) scoring significantly lower than those in the TAU condition (M = 22.67, SD 16.24) at T3 [d = 0.54, t(154) = 2.79; p = 0.006; with Holm adjustment p = 0.018].

Scheck et al.¹²¹ reported a significant effect for EMDR on depression [F(1,58) = 5.39; p = 0.024].

Shirk et al.¹¹⁷ reported significant reductions over time in BDI depression scores but no between-group differences.

Anxiety

Jensen et al.^112,113 found no main effect of treatment condition on children’s anxiety symptoms. Participants in the TF-CBT group [d = 0.30, t(150) = 1.47; p = 0.114; with Holm correction p = 0.114]. Analyses of the SCARED subscales found a main effect only for generalised anxiety disorder. Completer analyses produced similar results.

Scheck et al.¹²¹ reported a significant effect for EMDR on STATE anxiety [F(1,57) = 4.89; p = 0.031].

Behaviour

Rushton et al.¹¹⁶ found no significant differences in child problems between the two groups at 6 months’ follow-up, although a significant difference (p < 0.007) was found for ‘satisfaction with parenting’ in favour of the intervention group (effect size d = 0.7).

Jensen et al.^112,113 found a main effect of treatment condition on the SDQ (interpreted as general mental health problems). Participants in the TF-CBT group had significantly lower scores (M = 11.95, SD 6.51) than those in the TAU group (M = 14.54, SD 6.12) at the end of therapy [d = 0.45, t(152) = 2.46; p = 0.015; with Holm adjustment p = 0.030]. Completer analyses produced similar results.

In Linares 2012,¹¹⁵ the authors report that physical aggression decreased over time for both groups (IY, UC) but there were no between-group differences. After adjusting for gender, ethnicity, initial diagnosis of attention deficit hyperactivity disorder and study site, children in the UC group showed more improvement than those in the IY training group on foster parent reports of physical aggression. Rates of improvement were highest among children in the UC condition. Teachers reported no differences.

In Linares 2006,¹¹⁴ intervention children were reported as having lower CBC externalising T scores [F(1,97) = 2.71; p = 0.10] and Eyberg Child Behavior Inventory (ECBI) total T score [F(1,94) = 2.30; p = 0.13] at follow-up but these were not statistically significant.

Farkas et al.¹²⁰ reported significant differences on the CBCL²⁶⁰ in favour of MASTR/EMDR for externalising behaviour [F(1,40) = 9.77; p = 0.05], but not for internalising behaviour.

Risky behaviour

Champion 2012¹¹⁰ reported a significant impact of the intervention, with those who had received the theory-based (ARRM²⁰⁹) CBT intervention experiencing fewer infections at intervals of 0–6 months (0% vs. 6.6%; p = 0.001), 6–12 months (3.6% vs. 7.8%; p = 0.005, 95% CI 0.001 to 0.386) and 0–12 months (4.8% vs. 13.2%; p = 0.002, 95% CI 0.002 to 0.531).

Self-control

Linares 2012¹¹⁵ report a main effect in relation to foster care reports of higher levels, and steeper rates of improvement in relation to self-control, in favour of the control group (UC).

Self-esteem

Scheck et al.¹²¹ reported a significant effect of EMDR for the TSCC³³⁵ [F(1,57) = 4.573; p = 0.04].

Parent–child relationships

Emotional Freedom Questionnaire (EFQ) scores among control children in the Rushton et al. study¹¹⁶ remained unchanged, whereas they were more positive at all time points for the intervention group; however, these were not significant when controlling for baseline scores.

Although not directly relevant, Linares et al.¹¹⁴ reported a significant difference between the intervention and UC group on coparenting flexibility [F(1,104) = 4.14; p < 0.05], coparenting problem-solving [F(1,102) = 6.38, p < 0.01] and coparenting total [F(1,97) = 5.13; p < 0.05]. This was a key aim of the study¹¹⁴ and was likely to have a positive impact on the experience of children in foster care.

Summary

Four^{112,114–116} of the nine studies that we identified focused on helping caregivers (including birth parents, adoptive or foster parents) to address the consequences of maltreatment, by enhancing their knowledge and skills and providing support. Unsurprisingly, most of these studies provided services to carers of children aged < 10 years, but one¹¹² study recruited those caring for older children. The studies compared modifications of the IY parenting programme with either TAU¹¹⁴ or, in the only UK study,¹¹⁶ with an educational approach.

The other five^{110,111,117,120,121} studies provided services directly to maltreated young people aged > 12 years. They included two^120,121 studies that compared EMDR with alternative treatments (as usual), and three^110,111,117 studies comparing modified forms of CBT or CBT ‘plus’ with either no treatment, UC or a modified form of UC.

Overall, within the range of different studies included in this section, there was some reduction in symptoms PTSD and depression with treatment. Results regarding improvement in children’s behaviour vary between the studies. One¹¹⁵ study targeted physical aggression and self-control in foster children, and found no differences between children whose foster parents had participated in the IY programme and those who had received UC. Indeed, when appropriate adjustments were made, those in the control group did rather better than those in the experimental group. No differences were found between the behaviour of children whose adoptive parents received a CBT parenting programme based on IY, although these parents were significantly more satisfied with parenting and were less likely to use negative parenting approaches than those in the control group.¹¹⁶

Conflicts between foster parents and biological parents contribute to placement instability, and work against reunification. One¹¹⁴ study examined an intervention designed to improve relationships between parents, foster parents and children, and enhance the consistency of parenting across the two homes. The results of this study were very positive, and have relevance to the UK context.

One¹¹⁰ study, focusing specifically on risky behaviour by girls, showed reduction in rates of STIs.

It is difficult to draw conclusions about EMDR, as in one¹²⁰ study this was a very different intervention to the standard protocol, and it was compared with another treatment arm that was quite intensive.¹²⁰

Completeness and applicability

Only one study was conducted in the UK,¹¹⁶ and one study was conducted in Norway.¹¹² Both studies^112,116 focused on children in a family context, and aimed to improve outcomes for children who had experienced very serious maltreatment. In the UK study, participants were recruited from social welfare agencies that were responsible for supporting adoptive parents. In Norway, the settings were trauma clinics, that is, routine community settings.

Webster-Stratton’s IY programme has been endorsed as an evidence-based parenting programme in the UK, and the applications evaluated in three^114–116 of these studies have relevance to the needs of children in foster and adoptive care in the UK. The application of this programme to some of the challenges of fostering maltreated children is highly relevant, but the findings are mixed and the evidence base is limited.

Quality of the evidence

Most studies are small and there are few of them. Their heterogeneity (in terms of interventions, participants and outcome measures) prevented us from combining data in meta-analyses, and therefore the results from this part of the review should be treated with caution.

Overall, the quality of the evidence relating to studies of CBT for children who have experienced a range of forms of maltreatment is moderate, largely because of the impact of lack of information, which, if available, might demonstrate enhanced quality – or the reverse.

Study designs

The Becker-Weidman^134,135 study was a COS. The remaining studies^{122–133,136–140} were randomised trials.

Location

All studies were conducted in the USA^{123–129,131–136,140} except for Moss et al.¹³⁰ (Quebec, Canada), Hughes and Gottlieb¹³⁹ (Eastern Canada) and Thomas and Zimmer-Gembeck^137,138 (Australia).

Sample sizes

Controlled studies ranged from small to a moderately large sample size; the smallest sample size was 60 (with only 46 ultimately included in the analyses)^125,126 and the largest were 151^137,138 and 210.¹³¹ See Chapter 3 for further information.

Participants

Maltreatment

Interventions and comparisons

Number and duration of sessions

Attachment

Three^122,123,130 studies assessed the impact of the intervention on attachment using the Ainsworth Strange Situation Procedure.³³⁶

Dozier et al.^125,126 asked foster parents to record infants’ behaviour when distressed and in the presence of their primary caregiver using the Parent Attachment Diary (PAD⁶⁹⁷). Spieker et al.¹³¹ used the Toddler Attachment Sort-45,³⁵⁰ a modified version of the Attachment Q-sort⁶⁹⁸ to assess children’s attachment security, and Becker-Weidman^134,135 used the Randolph Attachment Disorder Questionnaire.³⁶⁴

Internal working models

Toth et al.¹³³ specifically explored the effectiveness of two developmentally informed preventative interventions on children’s internal representations of self, and of self in relation to other, using a narrative story stem task from the MacArthur Story Stem Battery (Bretherton I, Oppenheim D, Buchsbaum H, Emde RN & the MacArthur Narrative Group, University of Wisconsin-Madison, 1990, unpublished).

Child behaviour

Six^{130–132,134,135,137,138} studies examined the impact of attachment-based interventions on child behaviour, using the CBCL.²⁶⁰

In addition to the CBCL,²⁶⁰ Thomas and Zimmer-Gembeck 2011¹³⁷ and 2012¹³⁸ assessed the impact of PCIT on child behaviour using the ECBI (parent report),³¹⁰ and Spieker et al.¹³¹ did so using the Brief Infant Toddler Social and Emotional Assessment (BITSEA³⁵²) and selected items from the Bayley III Screening Test.³⁵⁵ Chaffin et al.¹³⁶ used the Behavior Assessment System for Children.³⁶⁵

Hughes and Gottlieb¹³⁹ used the Child Autonomy Observational Scale to assess child autonomy. Again, this tool was developed for the study, based on the theoretical underpinnings of Deci and Ryan.^699,700

Dozier et al.^125,126 assessed the impact of the intervention on children’s behaviour using the Parent Daily Report (adapted from Chamberlain and Reid⁴⁵⁸).

Child stress

Dozier et al.^125,126 and Cicchetti et al.¹²⁴ used analyses of cortisol to assess levels of stress in infant participants.

Parent behaviour

The Dyadic Parent–Child Interaction Coding System (DPICS)^701,702 was used by all three PCIT studies,^136–138 although Chaffin et al.¹³⁶ used the DPICS-II⁷⁰² and Thomas and Zimmer-Gembeck^137,138 used the DPICS-III.⁷⁰¹

Sprang¹³² and Thomas and Zimmer-Gembeck^137,138 examined the effect of the intervention on parents’ abuse potential using the Child Abuse Potential Inventory (CAPI³⁶⁶).

Parental stress

Spiekar et al.,¹³¹ Sprang¹³² and Thomas and Zimmer-Gembeck^137,138 assessed the impact of the intervention on parenting stress, using the Parenting Stress Index-Short Form.³⁴²

Maternal sensitivity

The primary caregiver outcome in the Spieker et al.¹³¹ study was maternal sensitivity, assessed using a modified score of the Nursing Assessment Teaching Scale,³⁵⁸ the secondary outcomes being a measure of parenting support for the child, using the Indicator of Parent–Child Interaction³⁵¹ and commitment to the child, assessed by answer to interview questions from This Is My Baby.³⁵⁹ Moss et al.¹³⁰ measured maternal sensitivity using the Maternal Sensitivity: Maternal Behaviour Q-Set.³³⁹

Parenting behaviour

In the Hughes and Gottlieb¹³⁹ study, a scale was developed that reflected the theoretical underpinnings of the study, namely the Parenting Skills Observation Scale (developed by the authors for this study).

Other

Valentino et al.¹⁴⁰ assessed changes in elaborative and emotion-rich reminiscing using video-taped and audio-taped conversations and the Peabody Picture Vocabulary Test⁷⁰³ to assess parents and children is receptive language.

Sequence generation

All studies, except for Spieker et al.¹³¹ and Hughes and Gottlieb,¹³⁹ were judged to be ‘unclear’ risk of bias for sequence generation because no information was provided on how the randomisation sequence was generated. Spieker et al.¹³¹ and Hughes and Gottlieb¹³⁹ were assessed as ‘low’ risk. Spieker¹³¹ stated that they used a computer-generated sequence, blocked by caregiver type. Hughes and Gottlieb¹³⁹ referred to a random numbers chart.

Allocation

All studies^122–140 were judged unclear risk of bias on allocation concealment because of a lack of information.

Blinding of personnel and participants

All but two^123,125,126 studies were assessed as unclear risk of bias for blinding of participants, predominantly due to lack of information. Dozier et al.^125,126 stated that ‘foster parents and birth parents were blind to condition’ and was therefore judged low risk of bias. Cicchetti 2006¹²³ did not blind participants but this was deemed unlikely to influence outcomes of this study because all participants in this study were infants aged, on average, 13–31 months.

Blinding of outcome assessors

All studies, except for Sprang,¹³² were judged as ‘low risk of bias’ for blinding of outcome assessors because assessors were blinded to the hypotheses, the assignment of participants and/or to the collected data. Thomas and Zimmer-Gembeck 2011¹³⁷ did not state that their outcome assessors were blinded, but this trial used the same procedures and outcomes measures as the 2012¹³⁸ trial to which it was linked, and in the 2012 study the outcome assessors are said to be ‘blind to treatment condition’. Sprang¹³² was assessed as high risk as the authors relied solely on self-completed measures collected by the treatment staff at post test.

Incomplete outcome data

With the exception of Toth et al.,¹³³ all studies were deemed to be at low risk of bias. Some studies^{122,136–138} had no missing data, some studies^{127–129,131,132} used a supplementary ITT analysis and the remaining studies^{123–126,130,133–135,139,140} conducted analyses of those who dropped out and those who completed treatment, and concluded that they did not differ. The Toth et al.¹³³ study was assessed as ‘unclear’ risk of bias as missing data were not reported and analyses included only participants who completed treatment.

Selective outcome reporting

Cicchetti 2006¹²³ was judged as ‘unclear’ risk of bias because, despite prospective registration and a (broad-brush) description of measures relevant to the domains outlined in that protocol, the paper reports only on attachment classifications. Cicchetti 2011¹²⁴ was judged as ‘unclear’ risk of selective outcome reporting bias because no means or SDs were presented and only latent growth curve data.

Spieker et al.¹³¹ and Lieberman et al.^127–129 also registered their trials (ClinicalTrials.gov identifiers NCT00339365 and NCT00187772, respectively). Spieker et al.¹³¹ reports on those outcomes listed in the trial registration and was judged as ‘low’ risk of bias on this domain. Lieberman et al.^127–129 reports on all primary outcomes and one of two secondary outcomes. Overall we judged this study to be ‘low risk’ of bias on this domain. The secondary outcome not reported on by Lieberman et al.^127–129 is child’s cognitive functioning.

All but one^125,126 of the remaining studies were deemed unclear risk of bias, as although they appeared to report on all of those outcomes expected, without access to the original study protocols, we cannot be certain. Dozier et al.^125,126 was assessed as ‘high risk of bias’ in light of the fact that two papers reporting the results of this study use different samples and report different outcomes without complete cross-referencing to a statement of all per-protocol outcomes.

Other sources of bias

No other potential sources of bias were identified in most of the studies. However, Sprang¹³² had an unclear risk-of-bias assessment because the authors reported that expectancy effects and trust in instructors may have played a role in the findings. Furthermore, Bernard et al.¹²² and Dozier et al.^125,126 had an additional high risk of bias because the Ainsworth Strange Situation Procedure was used as a measure for participants aged > 24 months, which extends beyond the age range for which it has been validated.

Summary details of the risk-of-bias assessments of these trials can be found in Appendix 10.

Attachment-focused interventions

Secure attachment

Bernard et al.,¹²² Cicchetti 2006¹²³ and Moss et al.¹³⁰ assessed the impact of attachment-based interventions on the security of a child’s attachment as measured by Ainsworth Strange Situation Procedure. The pooled estimate using a random-effects model was 0.14 (SMD) (95% CI 0.03 to 0.70) (Figure 17).

FIGURE 17

Secure attachment. df, degrees of freedom; M–H, Mantel–Haenszel.

The I²-statistic indicates that 82% of the variation in the point estimates is due to heterogeneity (p-value for heterogeneity = 0.004; τ² = 1.52).

Disorganised attachment

Bernard et al.,¹²² Cicchetti 2006¹²³ and Moss et al.¹³⁰ assessed the impact of attachment-based interventions on a reduction of a child’s disorganised attachment style, as measured by Ainsworth Strange Situation Procedure. The pooled estimate using a random-effects model was 0.23 (SMD) (95% CI 0.13 to 0.42; p = 0.00001) (Figure 18). The I²-statistic indicates 17% of the variation in the point estimates is due to heterogeneity (p-value for heterogeneity = 0.030; τ² = 0.05).

FIGURE 18

Disorganised attachment. df, degrees of freedom; M–H, Mantel–Haenszel.

Avoidant attachment

Both Cicchetti 2006¹²³ and Moss et al.¹³⁰ assessed the impact of attachment-based interventions on a child’s avoidant attachment style, as measured by Ainsworth Strange Situation Procedure. The pooled estimate using a random-effects model was 0.90 (SMD) (95% CI 0.13 to 6.37; p = 0.09) (Figure 19).

FIGURE 19

Avoidant attachment. df, degrees of freedom; M–H, Mantel–Haenszel.

The I²-statistic indicates 64% of the variation in the point estimates is due to heterogeneity (p-value for heterogeneity= 0.09; τ² = 1.29).

Parent–child interaction therapy

Child behaviour (externalising)

Both Chaffin et al.¹³⁶ and Thomas and Zimmer-Gembeck 2011¹³⁷ and 2012¹³⁸ assessed the impact of PCIT on child externalising behaviours (as measured by various scales). Thomas and Zimmer-Gembeck 2012¹³⁸ used a control group that overlapped with their 2011 study,¹³⁷ and the extent of the overlap was unclear, so we decided only to include the data from their 2011 report of TV/PCIT in the meta-analysis to avoid the problem of double counting.

The pooled estimate, using a random-effects model, was 0.03 (SMD) (95% CI –0.38 to 0.43) (Figure 20). The I²-statistic indicates that 44% of the variation in point estimates between the two studies is due to heterogeneity, making it difficult to draw conclusions about the effect of PCIT on externalising behaviours/symptoms in maltreated children from these two studies. It is perhaps worth noting that Thomas and Zimmer-Gembeck 2012¹³⁸ also report significant improvements in the behavior of children whose parents participated in S/PCIT and, in both studies,^137,138 gains were said to be greater for those who completed PCIT.

FIGURE 20

Child externalising behaviour. df, degrees of freedom; IV, instrumental variable.

It is also worth noting that Thomas and Zimmer-Gembeck 2012¹³⁸ report a greater improvement in externalising behaviours/symptoms (p < 0.001) after treatment with S/PCIT than with wait-list controls. This study also found a greater improvement in externalising behaviours/symptoms (p = 0.002) after treatment with S/PCIT than with TV/PCIT.

Thomas and Zimmer-Gembeck 2011¹³⁷ and 2012¹³⁸ also measured the intensity of behaviour problems using the ECBI.

Thomas 2011¹³⁷ found a greater reduction in child behaviour problems (p < 0.001) and intensity (p < 0.001) after treatment with TV/PCIT than with wait-list controls.

Thomas 2012¹³⁸ found a greater reduction in child behaviour problems (p < 0.000) and intensity (p = 0.019) after treatment with S/PCIT than with wait-list controls.

Thomas 2012¹³⁸ also found a greater reduction in child behaviour problems (p = 0.001) after treatment with S/PCIT than with TV/PCIT. However, there was no difference between S/PCIT and TV/PCIT for child behavior intensity (p = 0.096).

Parenting interventions

In the study by Valentino et al.,¹⁴⁰ children in the intervention group had richer memory recall (p < 0.01, d = 0.71) and made more emotion references (p < 0.001, d = 1.35) than control children during conversation with parents but not with experimenters.

Contrary to expectation, there was no effect of the Webster-Stratton IY parent programme on child autonomy in the Hughes and Gottlieb study.¹³⁹

Summary

We identified three groups of RBIs.

The first group of 10^122–135 studies addressed the problems of insecure or disorganised attachments among maltreated children. Of these, four^{122,125,126,132} studies evaluated the effectiveness of the short intervention ABC or based on ABC, and four^{123,124,127–129,133} studies evaluated an attachment theory informed intervention known as IPP. All were short-term, manualised programmes, and they focused on promoting sensitive and responsive care to children adversely affected by maltreatment, with the focus on the carer–child relationship and patterns of interaction. In only one^134,135 study were children and young people seen by a therapist. In this study,^134,135 the therapy, DDP, is described as a FT treatment based on attachment theory, in which the relationship between therapist and child, caregiver and child, and therapist and caregiver provide the context for treatment. The last¹³¹ study evaluated the impact of training early years professionals.

The body of evidence relating to attachment informed interventions is promising, particularly in relation to ABC and IPP interventions. Meta-analyses of data from three^122,123,130 studies (two ABC, one IPP) indicate significant increases in attachment security and decreases in disorganised attachment. These results are consistent with those reported in studies, the data for which we were unable to combine in these meta-analyses, with some indication that children in the control group evidenced deterioration, that is, more developed disorganised attachment. This suggests that timely interventions may be able to prevent cumulative consequences of maltreatment.

Findings in relation to children’s behaviour were mixed, and generally did not reach the level of statistical significance. One¹³⁰ study that recruited children with a wide age range (1–5 years) explored the moderating impact of age and found that reductions in child externalising and internalising problems was associated with increasing age in the intervention group, whereas a marginal increase in behaviour problems was found with age for the control group. Such analyses can only be hypothesis generating and, given the size of the study,¹³⁰ probably highly speculative. However, the authors note that the maladaptive trajectories for maltreated children increasingly diverge from their non-maltreated peers over time, with the transition to school often being particularly difficult for maltreated children who maintain increasingly socially dysfunctional patterns of aggression and social withdrawal. Therefore, an intervention that has the potential to reduce behaviour problems in preschool children may well be particularly helpful. A similar pattern was noted by Dozier et al.^125,126

Children’s stress levels, as measured by cortisol secretion patterns, also improved with attachment-based interventions.

We identified three randomised trials of PCIT. All were concerned with addressing behaviour problems resulting from physical abuse by helping parents change the way they interacted with their children. A meta-analysis of data from measures of child externalising behaviour indicated no effect of PCIT. Chaffin et al.¹³⁶ say in the discussion that this may be because, in this study, PCIT was evaluated as a parent treatment; the study included children older than those customarily included in PCIT for child behaviour problems. Given the relational context of maltreatment, and the emphasis placed by these authors on the importance of the escalating coercive interactions, this account reads as a ‘post hoc’ explanation of an unexpected finding. There is some uncertain evidence regarding the potential reduction in child externalising behaviour problems following PCIT. Thomas and Zimmer-Gembeck¹³⁷ report a decrease in the severity of behaviour problems of children receiving PCIT, as perceived by parents, but these children evidence no diminution in externalising behaviour as assessed by the CBCL²⁶⁰ (parent report).

As intended, the RET intervention improved children’s memory recall and emotion referencing, with their parents.

Completeness and applicability

None of these studies was conducted in the UK, but all are concerned with relationship problems that are familiar to mental health professionals and social workers. It is difficult to overstate the importance of secure attachment as a public health issue, and the attachment difficulties that so often result from maltreatment present a cumulative threat to children who experience abuse and neglect. The attachment interventions evaluated are relatively brief, manualised interventions that include techniques and strategies currently in use by many health professionals in the UK, for example video interactive guidance. Given the findings reported, well-conducted trials of these interventions in the UK would be appropriate prior to recommending their widespread adoption.

There is currently no strong evidence to support the use of either PCIT or Webster-Stratton’s IY Program as an intervention to address the emotional or behavioural problems of maltreated children, and their impact on improving parenting is clinically questionable.

Quality of the evidence

Only three^{122,125,126,132} of these studies were judged to be of high risk of bias in any of the seven domains assessed, but most were so poorly reported that judgements of unclear risk of bias outnumbered those of low risk. Overall, these studies do better in terms of detection bias (the blinding of outcome assessors) and attrition bias (incomplete outcome data). Six^{122–128,132} of these studies were conducted by two American teams, and essentially replicate one another, with variations in the profile of participants. Each team is evaluating an intervention that they have helped to develop. Additional, independent evaluations of both interventions (ABC and IPP) would be helpful, together with better reporting.

Study design

Of the eight^{107,108,141–144,147–150} controlled studies, six^{107,108,141–144,148,149} were randomised trials. One was the QEx Bagley and LaChance study¹⁵⁰ and one was a COS.¹⁴⁷

Location

One¹⁵⁰ study was conducted in in Canada. The remainder of the studies^{107,108,141–144,148,149} were carried out in the USA.

Sample sizes

Sample sizes were generally small. Linares 2015¹⁴¹ randomised 22 sibling pairs. Meezan and O’Keefe^148,149 and Swenson et al.¹⁴⁴ had samples of 81 and 90, respectively, and the remaining RCTs had sample sizes of 30¹⁴³, 43¹⁴² and 55^107,108, respectively. Bagley and LaChance¹⁵⁰ had a sample size of just 65 (after attrition and exclusions) and Schaeffer et al.¹⁴⁷ had a sample size of 25 youth–mother dyads.

Participants

Referrals: all systemic interventions

Linares et al.¹⁴¹ recruited eligible children and foster carers from three participating fostering agencies. Families in the other seven^{107,108,142–144,147–150} systemic intervention studies were recruited from CPS or similar governmental agencies. In the study by Danielson et al.,¹⁴³ families were also referred or recruited from clinics/treatment providers.

Maltreatment type

Sources of maltreatment

Where specified, the source of maltreatment was within the family, including biological parents, step-parents, or a parent’s cohabiting partner. In Danielson et al.¹⁴³ the source of maltreatment was not reported.

Interventions and comparisons

Number and duration of sessions

Post-traumatic stress disorder

Swenson et al.¹⁴⁴ and Schaeffer et al.¹⁴⁷ both used the TSCC.^378,383 Danielson et al.⁷⁰⁵ used the University of California Los Angeles PTSD Index for DSM-IV (Adolescent version) (UCLA-A) and Caregiver version.³⁷⁴

Depression

Danielson et al.¹⁴³ and Kolko^107,108 both assessed the impact of intervention on depression using the CDI.²⁹⁹

Bagley and LaChance¹⁵⁰ used the Center for Epidemiologic Studies Depression Scale (CES-D³⁶⁹).

Behaviour problems

Kolko 1996,^107,108 Swenson et al.¹⁴⁴ and Meezan and O’Keefe^148,149 assessed changes in problematic behaviour using various reporting forms and versions of the CBCL YSR.^{198,269,706,707} Danielson et al.¹⁴³ used the Behavior Assessment System for Children-Second Edition³⁶⁵ and Bagley and LaChance¹⁵⁰ relied on parent, social worker and self-report of problem/delinquent behaviours. Brunk et al.¹⁴² used the Behavior Problem Checklist (BPC; Quay and Peterson, University of Miami, Coral Gables, FL, 1975, unpublished).

Kolko^107,108 also assessed child conflict with the CCI⁶⁸⁸ and hostility using the Children’s Hostility Inventory.³⁰¹

Self-esteem

Bagley and LaChance¹⁵⁰ used the Rosenberg Self-Esteem Scale (RSES no reference provided by the authors).

Child functioning

Kolko 1996^107,108 assessed overall child functioning with the Kiddie Global Assessment Scale (KGAS) (no reference provided by the author) and assessed peer relationships using The Friendship Questionnaire.³⁰⁰ Meezan and O’Keefe^148,149 used the Children’s Action Tendency Scale CAS³⁸⁵ and the Index of Peer Relations³⁸⁶.

Substance use

Danielson et al.¹⁴³ used the Time Line Follow Back Interview³⁷⁶ and urine drug screens to assess the impact of intervention on substance use and risky behaviour.

Other outcomes measured by studies of systemic interventions

Sequence generation

Risk of bias in sequence generation was judged low in three trials: Kolko^107,108 used a computer-generated procedure based on Efron’s biased coin toss; Swenson et al.¹⁴⁴ used a computer-generated table of random numbers; and Danielson et al.¹⁴³ randomised participants using computerised blocked randomisation.

The remaining RCTs^{141,142,148,149} were judged to be ‘unclear’, as the only information available was that participants were ‘randomised’.

Allocation concealment

None of the RCTs included provided adequate information on allocation concealment and so all were judged as being of unclear risk of bias.

Blinding of participants and personnel

With the exception of Swenson et al.,¹⁴⁴ who stated clearly that participants were not blinded, no study referred to the blinding of participants or personnel. Given the nature of the intervention, it is unlikely that blinding was possible in any of the studies, so we judged these to be high risk of bias.

Blinding of outcome assessors

Blinding of outcome assessors was carried out in the studies by Brunk et al.¹⁴² and Linares et al.,¹⁴¹ so these studies were classed as being of low risk of bias. The use of self- and parent-report measures in the studies by Danielson et al.,¹⁴³ Meezan 1998^148,149 and Swenson et al.¹⁴⁴ resulted in a judgement of ‘high risk of bias’, and also in Kolko 1996^107,108 in which self-report measures undertaken at the beginning of the FT sessions with the family all present as a potential source of bias.

Incomplete outcome data

We judged the risk of bias as a result of missing data to be low in three trials.^141,143,144 These trials minimised potential bias by maintaining a high retention rate and clear reporting¹⁴⁴ or using ITT analysis.^141,143 Attrition in Kolko^107,108 was around 6% (one case) in the FT arm, 20% (five cases) in the CBT arm and 17% (two cases) in the routine community services groups. The differential attrition between the two arms resulted in a judgement of high risk of bias (reasons for dropout might be related to the intervention). Brunk et al.¹⁴² and Meezan and O’Keefe^148,149 were also judged to be of high risk of bias. Meezan and O’Keefe^148,149 took no account of attrition and analyses were based on available case data, resulting in a judgement of high risk of bias. Dropouts in Brunk et al.¹⁴² were evenly distributed across the two arms of the trial, but reasons for drop out are not given, and the authors do not report main effects. We therefore deemed this trial high risk of bias.

Selective outcome reporting

Three^{142,144,148,149} studies were assessed as being of high risk of bias: Swenson et al.¹⁴⁴ did not report means and SDs for pre and post measures at each time point, and effect sizes were missing for some measures, leading to high risk of bias; Brunk et al.¹⁴² do not report data on main effects or provide post-treatment means, with insufficient data for effect size calculations, and, although stating that they measured child behaviour problems using the BPC (Quay and Peterson, 1975, unpublished), they present no results for this measure. Meezan and O’Keefe^148,149 did not report data for ‘non-significant’ results, leading to high risk of bias.

Linares 2015¹⁴¹ was assessed as unclear, as, although there is no evidence that other outcomes were planned and then not reported, there is no published protocol for this study.

The study by Danielson et al.¹⁴³ was judged to be of ‘low risk of bias’: the authors report on the primary outcome measure specified in the trial registration (ClinicalTrials.gov NCT00998153) and two of the three secondary measures of PTSD and family environment, but not on risk behaviours as measure by the Youth Risk Behavior Survey.

Results: systemic interventions

Meta-analysis was not possible across these studies, which are summarised narratively. Given the heterogeneity across studies, we report the results in the three groupings used above: FT, MST and ‘other systemic’.

Family therapy: results

The following results are these reported by Kolko.^107,108

Multisystemic therapy: results

No meta-analysis was possible for MST because of the limited number of studies reporting appropriate data. For these interventions, we report the results for those outcomes that were directly related to children and young people, and not those related to parents, which includes measures of parental substance abuse,¹⁴⁷ parenting stress,¹⁴² parental mental health,^142,144,147 parenting behaviour,^144,147 social support¹⁴⁴ and re-abuse.¹⁴⁴

Family-based systemic interventions (including transtheoretical): results

Summary

We identified eight^{107,108,141–144,147–150} studies that evaluated a heterogeneous group of interventions informed by systems theory and offered to different participants. Four studies evaluated various forms of MST, comparing this with CBT,¹⁴² TAU,¹⁴³ enhanced outpatient treatment,¹⁴⁴ and Comprehensive Community Treatment.¹⁴⁷

Of the remaining four studies, one compared systemic FT with CBT;^107,108 one worked systemically with families in which a young person had been the subject of incestuous abuse;¹⁵⁰ a third used MFGT;^148,149 and the final study¹⁴¹ described itself as a transtheoretical intervention that focuses on three family subsystems: sibling pairs in foster care, the foster parent, and foster parent/sibling pairs.

All of these interventions included cognitive–behavioural strategies and psychoeducation, but their underlying theories of change were primarily systemic.

The four^{142–144,147} MST studies are heterogeneous and the results are variable. Only one¹⁴⁴ study of the three^143,144,147 MST studies assessing the impact of the intervention on PTSD reported a significant benefit in favour of MST. This rather larger study¹⁴⁴ (n = 90) halved the percentage of youth scoring in the clinical range for self-reported PTSD symptoms from 18% at baseline to 9% at 16 months post baseline, in contrast with the group receiving enhanced outpatient services, for which the percentage increased from 19% to 21%. Retention in both the treatment and the study¹⁴⁴ was high, and the intervention appears to have been successful at not only addressing the mental health symptoms of participating youth, but also addressing those aspects of parenting associated with maltreatment from both youth and parent perspectives, in particular reducing parental neglect and assault. The other two studies^143,147 found no significant between-group differences in depression or PTSD. Schaeffer et al.¹⁴⁷ observe that these young people reported subclinical levels of concern at baseline. Danielson et al.¹⁴³ (who studied sexually abused children) hypothesise that the absence of effect is, in part, because of the small size of the study and considerable baseline inequality. In this study¹⁴³ the primary outcome was substance use risk and mental health problems, and the authors report a significant reduction in substance use and associated risk factors among MST youth compared with TAU.

As indicated earlier, both CBT and FT outperformed routine community services. This early comparative study^107,108 was one of the first to submit FT to rigorous evaluation and the authors note that FT has been less frequently used in the context of child maltreatment than CBT; they recommend further development, including the incorporation of a structured approach to address the parent–child relationship. At the same time, they highlight the need for CBT to give attention to how best to discuss children’s attributions of their victimisation.

There is no strong evidence of benefit to children from multigroup FT with abusive and neglectful families.^148,149 Linares et al.¹⁴¹ report promising, but mixed results of an intervention designed to reduce sibling aggression in foster care, and this small study¹⁴¹ requires replication before any conclusions can be drawn. However, there is some indication that foster parents can learn strategies to minimise sibling aggression and its adverse effects on psychological child well-being.⁷¹²

Completeness and applicability

All of these generally very small studies were conducted in North America. MST enjoys a strong evidence base in relation to vulnerable groups such as young offenders, and the extension to abusive and neglectful families suggests that MST-CAN has potential as an effective intervention for families whose behaviour places their children at risk of significant harm. Results of the other MST studies are less persuasive than those of Swenson et al.¹⁴⁴ but of particular interest is the QEx study by Schaeffer et al.¹⁴⁷ This study¹⁴⁷ examined the impact of an MST intervention for maltreating families in which parental substance abuse was a major issue. Although this study¹⁴⁷ showed a reduction only in child anxiety, given the numbers of children with child protection plans for whom parental substance misuse is a key factor, this intervention is potentially highly relevant, and the authors report an ongoing, large-scale longitudinal randomised trial that will allow the investigation of the programme’s effectiveness with different subgroups, that is, different substance misuse and maltreatment characteristics. MST is a costly intervention, but may be cost advantageous.

Quality of the evidence

Six^{107,108,141–144,148-149} of the included controlled studies were randomised trials, but the majority of risk-of-bias judgements made were ‘unclear’ as a result of poor reporting. The QEx study¹⁴⁷ was judged good overall (but it was a small study), and the COS¹⁴⁷ was judged adequate. Overall, the quality and extent of the evidence for systemic interventions to address the effects of child maltreatment are parsimonious.

Study design

Seven^151–157 of the psychoeducational intervention studies were randomised trials. Trowell et al.¹⁵⁵ was designed as an other-treatment control study, comparing a psychoeducational group with IT.

Three^158–160 studies used a QEx design and the remaining seven^161–168 studies were COSs.

Location

All^{151,152,154,156,157} but four^{155,158,159,165} of the psychoeducational intervention studies were conducted in North America.

One¹⁵⁵ RCT was conducted in the UK. The remaining six^{151,152,154,156,157} randomised trials took place in the USA^151–154 and Canada.^156,157

Two^159,160 of the three QEx studies were carried out in Canada and the third¹⁵⁸ was conducted in the Netherlands.

The remaining COSs took place in the USA^161,162 Canada^{163,164,166–168} and Spain.¹⁶⁵

Sample sizes

All of the randomised trials made use of small to moderate sample sizes, ranging from a total of 42 participants (with only 38 participants completing post-treatment interviews)¹⁵⁶ to 181 participants (with 174 participants completing the follow-up interview).¹⁵¹

Of the QEx studies, Noether et al.¹⁵⁸ had a large sample size of 253 participants (with 210 participants completing the 12-month follow-up interview); Simoneau et al.¹⁵⁹ and Tourigny 2007¹⁶⁰ recruited 49 and 55 participants, respectively.

Sample size in four of the COSs was small, ranging from 27 participants¹⁶¹ to 42.^166,167 Holland et al.¹⁶⁴ and Hébert et al.¹⁶³ had samples of 66 and 90, respectively, although data were presented for only 88 participants in the Hébert and Tourigny¹⁶³ study, and Duffany and Panos¹⁶² recruited a sample of 617.

Participants

Interventions and comparisons

All 17^151–168 studies used a group-based format of psychoeducation. Although the interventions were predominantly child focused, one intervention provided training for foster carers.¹⁶¹ Control group participants in all but two of the psychoeducational studies received either no intervention or management as usual. Overbeek et al.¹⁵³ made use of a control programme ‘Jij hoort erbij’ (‘You belong’), based on an analysis of non-specific factors used in the specific factors intervention programme. Trowell et al.¹⁵⁵ compared group psychoeducation with brief, focused, individual psychoanalytic psychotherapy.

Number and duration of interventions

Duration of sessions ranged from 1 to 2 hours each, with sessions lasting 9–20 weeks (most are between 10 and 12 weeks), with the exception of Santibáñez,¹⁶⁵ in which the programme lasted 4 months, and appears to have comprised weekly group and individual sessions plus everyday activities in the residential homes to promote generalisation of learning.

Post-traumatic stress disorder

Four measures of children’s post-traumatic stress symptoms were used across the seven^{153,155,160,163,166–168} studies assessing this outcome. Four^{153,160,166–168} studies used the Trauma Symptom Checklist (TSC). Overbeek et al.¹⁵³ also used the Trauma Symptom Checklist for Young Children (TSCYC), Hébert and Tourigny¹⁶³ used the Children’s Impact of Traumatic Events Scale-II (Wolfe, unpublished assessment instrument – available from VV Wolfe, Child and Adolescent Centre, London Health Sciences Centre, London, ON, Canada) and Trowell et al.¹⁵⁵ used the 1989 version of Orvaschel’s PTSD scale⁴⁰⁶ – an extension of the KSADS instrument.²⁶⁸ The Trauma Symptom Checklist-40 (TSC-40⁴⁰⁸) was used by Wolfe et al.¹⁵⁷ to assess symptoms of emotional distress.

Depression

Hébert and Tourigny¹⁶³ assessed the impact of psychoeducation on childhood depression using the CDI.⁸⁰ Trowell et al.¹⁵⁵ used a shortened version of the KSADS.²⁶⁸

Self-harm

Self-harm was assessed in all three Tourigny studies^{160,166–168} using the Self-Injurious Behavior Questionnaire (Sadvosky, unpublished). Holland et al.¹⁶⁴ also examined this outcome using administrative data.

Anxiety

Hébert and Tourigny et al.¹⁶³ assessed anxiety as an outcome, using the RCMAS.²⁵⁶ This study also assessed symptoms of dissociation using the Child Dissociative Checklist.⁴²⁶

Behaviour problems

The effect of psychoeducation on children’s problem behaviour was examined by seven studies using various versions of the CBCL. The studies by Graham-Bermann et al.,¹⁵¹ Overbeek et al.,¹⁵³ Hébert and Tourigny¹⁶³ and Barth et al.¹⁶¹ used the Parent Report Form, and both studies by Tourigny et al.^{160,166–168} used the YSR Form.

Two studies assessed delinquency: Tourigny 2005^166,167 used the Criminal and Delinquent Behaviours Questionnaire⁷¹⁴ and Holland et al.¹⁶⁴ used administrative data.

Antisocial and criminal behaviour was assessed using the Antisocial and Criminal Behavior Questionnaire⁷¹⁵ in the study conducted by Santibáñez,¹⁶⁵ who also investigated cognitive mediators of aggression using a 20-item Likert-type scale.⁴²⁷

Noether et al.¹⁵⁸ used mother/carer ratings of the Behavioral and Emotional Rating Scale (BERS⁴¹⁰) as the measure of the primary outcome measure.

The primary outcome in Duffany and Panos¹⁶² was recidivism (being re-abused or becoming abusers), assessed using the Youth Outcome Questionnaire.⁴²⁴

Self-control

Self-control was assessed in Santibáñez¹⁶⁵ using the Shapiro Control Inventory⁴²⁸ in its Spanish version.

Social competence

In the Howell et al.¹⁵² study the primary outcome was enhancing social competence in children who had witnessed IPV which they measured using the Social Competence Scale (Conduct Problems Prevention Research Group³⁹⁵).

Self-concept and self-adequacy

Sullivan et al.¹⁵⁴ measured changes in the well-being of children who had been exposed to domestic violence using Harter’s Self-Perception Profile for Children (SPPC; 8- to 12-year-old version⁴³¹). Hébert and Tourigny¹⁶³ used the same measure as part of his larger battery of outcome measures for a group of sexually abused children.

All three Tourigny et al.^{160,166–168} studies assessed the impact of intervention on coping strategies using a French version of the Ways of Coping Questionnaire.⁴¹⁹ They also deployed four of the five dimensions that make up the Empowerment Scale,⁴²¹ namely optimism, self-efficacy, helplessness and justified anger plus the French version of the Children’s Attributions and Perceptions Scale.²⁷⁰

Sexual behaviour

Barth et al.¹⁶¹ and Holland et al.¹⁶⁴ examined the impact of intervention on children’s sexual behaviours using the CSBI.⁴²³ Holland et al.¹⁶⁴ also used administrative data.

Relationships

Healthy relationship skills were assessed in the study by Wolfe et al.¹⁵⁷ using the Adolescent Interpersonal Competence Questionnaire.⁴⁰⁹

Two studies^151,156 focused on children’s attitudes and beliefs about the acceptability of family violence. Graham-Bermann et al.¹⁵¹ assessed change in this outcome using the Attitudes About Family Violence scale,³⁹⁴ whereas Wagar and Rodway¹⁵⁶ used a Child Witness to Violence Questionnaire (no reference provided) to assess children’s knowledge of wife abuse, who children feel are responsible, and their responses and attitudes to anger, their problem-solving abilities related to safety skills.

Global functioning

Trowell et al.¹⁵⁵ measured social, psychological and school functioning using the KGAS (1986 version⁴⁰⁵), based on the Children’s Global Assessment Scale (CGAS).²⁶¹

Sequence generation and allocation concealment

It was not possible to judge whether sequence generation was adequately conducted, as five^153–157 of the seven trials provided no other information other than participants were ‘randomly assigned’. These studies were therefore assessed as ‘unclear’ risk of bias for both sequence generation and allocation concealment.

Graham-Bermann et al.¹⁵¹ and Howell et al.¹⁵² both describe a modified, sequential random assignment procedure. In the study by Graham-Bermann et al.,¹⁵¹ the first seven children were assigned to the ‘child-only intervention’, the next seven children to the ‘child plus mother intervention’ and the next seven children to the wait-list control. In the Howell et al.¹⁵² study, the first five families were allocated to the experimental arm and the next five to the control arm, but the paper said nothing about allocation concealment. Both studies were assessed as ‘low risk of bias’ for sequence generation and unclear for allocation concealment.

Blinding of participants and personnel

Trowell et al.¹⁵⁵ stated that it was not possible to blind participants. None of the other studies makes any reference to procedures to blind the participants or personnel, and neither is likely to have been undertaken, given the nature of the intervention. All studies were therefore assessed as ‘high risk of bias’ for this bias domain.

Blinding of outcome assessors

Outcome assessors were blinded in Overbeek et al.¹⁵³ Graham Bermann et al.¹⁵¹ state that baseline interviews were conducted by researchers blind to group assignment and separate from those who provided the intervention; however, with the exception of child attitudes, the study¹⁵¹ relied solely on mother self-report and we judged this study¹⁵¹ overall to be of high risk of bias on this domain.

Wagar and Rodway¹⁵⁶ state that group leaders conducted the pre-group interviews and administered the measurement tools, as well as facilitating the groups, and so this study was judged to be of high risk of bias for outcome assessment. Trowell et al.¹⁵⁵ state that blinding of assessors was probably compromised because the children and their mothers often mentioned the specific therapy during assessment.

The absence of information on blinding of outcome assessors, together with the use of use of self-report and parent-reported measures in the studies by Howell et al.,¹⁵² Sullivan et al.¹⁵⁴ and Wolfe et al.,¹⁵⁷ suggest that outcome assessors were not masked in these studies, leading to a judgement of high risk of bias.

Incomplete outcome data

Three^153–155 of the seven trials sought to minimise potential attrition bias by analysing participants based on ‘assigned treatment’ rather than ‘treatment completed’ and so were judged low risk of bias. Graham-Bermann et al.¹⁵¹ identify differential attrition as a cause of concern and this study was accordingly judged high risk of bias, as were the studies by Wagar and Rodway¹⁵⁶ (which lost around 10% of children, all four from the experimental arm) and Wolfe et al.¹⁵⁷ (which reported differential attrition of 21% experimental and 11% control and undertook completer-only analyses). Howell et al.¹⁵² was also deemed high risk of bias because 21 of the 113 children recruited to the study were missing from the final analyses (seven dropouts in the experimental group and 14 in the control group).

Selective outcome reporting

Although the studies all differed somewhat in their choice of outcomes, there appears to be no evidence that other outcomes were planned and then omitted from the results. In the absence of study protocols it is extremely difficult to assess the risk of selective outcome reporting. In that respect, almost all were assessed as unclear risk of bias. However, two^156,157 of the included studies failed to report specific means and SDs for their post-intervention results and were therefore judged to be of high risk of bias.

Other sources of bias

Some other potential sources of bias were noted in two^151,154 of the studies, including a suggestion that the sample was nationally unrepresentative¹⁵¹ and potential bias associated with paying participants for their continued involvement in the study.¹⁵⁴

Full details of risk-of-bias assessments for each study can be found in Figures 33 and 34 in Appendix 10.

Post-traumatic stress

Both Wolfe et al.¹⁵⁷ and Overbeek et al.¹⁵³ assessed the impact of the psychoeducational intervention on children’s post-traumatic stress, albeit with two measures (as measured by the TSC-40⁴⁰⁸ and the TSCYC,³⁹⁹ respectively). We were unable to combine these data in a meta-analysis. Based on ITT and completer analyses, Overbeek et al.¹⁵³ found no differences between children in the experimental and control group; children in both groups improved.

Working with teenagers with histories of child maltreatment, Wolfe et al.¹⁵⁷ reported improvements in symptoms of trauma.

Children’s PTSD was also assessed by Trowell et al.¹⁵⁵ However, unlike the studies by Overbeek et al.¹⁵³ and Wolfe et al.,¹⁵⁷ this study¹⁵⁵ compared two active treatments. In Trowell et al.,¹⁵⁵ univariate analyses failed to identify a difference between those receiving group or IT on the PTSD dimension of ‘persistent symptoms of increased arousal’ used in the study.⁴⁰⁵ Following Cohen,⁷¹⁶ the authors used an effect size of 0.5 as a threshold of moderate effect, and undertook no further analyses relating to PTSD symptoms.

Controlling for the impairment score on the KGAS⁴⁰⁵ in a multivariate analysis, the authors report a significant effect of IT for ‘the re-experiencing of trauma’ dimension of PTSD (baseline to 1- and 2-year follow-up, and baseline to exit), and – for the ‘persistence/avoidance of stimuli’ dimension – at baseline to first-year follow-up. When KGAS is replaced by the baseline score on the same dimensions, the significance of the effects is attenuated.

Children’s PTSD symptoms were also assessed in three COSs^{163,166–168} and one QEx study.¹⁶⁰ Owing to the high risk of bias in all four of these studies,^{160,163,166–168} results were not incorporated into the meta-analyses and are instead presented narratively. Their results indicate that adolescents who were part of the psychoeducational intervention group improved significantly compared with controls on post-traumatic stress scores at both post test^{160,163,166,167} and 6-month follow-up.^166,167 A fourth study¹⁶⁸ found no statistically significant differences, although clinical measures suggested improvement for the treatment group compared with the control group.

Depression

In the trial conducted by Trowell et al.,¹⁵⁵ the between-group differences on the KGAS failed to reach the threshold effect size adopted by the authors of 0.5 (following Cohen⁷¹⁶). In the one CS¹⁶³ that assessed childhood depression, participants in the intervention group demonstrated only marginally fewer symptoms of depression following the intervention.

Behavioural problems

Graham-Bermann et al.¹⁵¹ and Overbeek et al.¹⁵³ assessed the impact of the psychoeducational intervention on child externalising and child internalising behaviours (as measured by the CBCL).

Externalising behaviours The pooled estimate using a random-effects model was –0.19 (SMD) (95% CI –0.45 to 0.06) (Figure 21). The I²-statistic indicates 0% of the variation in the point estimates is due to heterogeneity.

FIGURE 21

Externalising behaviours. df, degrees of freedom; IV, instrumental variable.

Internalising behaviours The pooled estimate using a random-effects model was –0.00 (SMD) (95% CI –0.25 to 0.25; p = 0.84) (Figure 22). The I²-statistic indicates 0% of the variation in the point estimates is due to heterogeneity.

FIGURE 22

Internalising behaviours. df, degrees of freedom; IV, instrumental variable.

Children’s behavioural problems were also assessed in three^{161,163,166–168} COSs and two^158,160 QEx studies.

Four^{158,163,164,166,167} of the included studies found that post-test scores on behavioural measures were significantly lower for children who had received a psychoeducational intervention.

One study¹⁶⁰ found that although intervention group participants (all girls) showed a significant decrease in internalising behaviours and social problems, change scores on externalising behaviours problems were not significant. One study¹⁶⁸ found no statistically significant differences for internalising or externalising behaviours. Another study¹⁶² found that approximately one-third (15/47) of the children showed no change, or an insignificant worsening of behavioural and/or emotional symptoms following the intervention, and the last study¹⁶¹ reported that behaviour in both intervention and control groups worsened in equal measure from the pre-test to the follow-up.

Other outcomes

Summary

We identified a large number of studies of psychoeducational interventions including seven^151–157 randomised trials, three^158–160 QEx studies and seven^161–168 COSs. All included or comprised a group-based format, and although the groups used a variety of formats (activities, discussion, etc.) they generally included an educative component (focused on the nature of maltreatment suffered by the children), affect modulation, emotion regulation skills, coping and processing, social skills and addressing future safety.

Although the findings of this heterogeneous body of evidence vary somewhat, there is evidence of effectiveness in relation to symptoms of PTSD.

The evidence for improving children’s behaviour is more mixed. On the basis of ‘vote counting’ most report a positive impact on externalising, internalising, delinquency and antisocial behaviour and self-injurious behaviour. A minority report ‘no difference’ and Barth et al.¹⁶¹ report a worsening in children’s behaviour. Participants in the Barth et al.¹⁶¹ study were foster parents caring for sexually abused children, who rarely implemented the homework tasks required of them during this short intervention but who appreciated the intervention.

Where examined, similar results in favour of psychoeducation were reported for positive changes, such as improvements in coping, enhanced self-competence, social competence and sense of empowerment. Generally, the involvement of parents (often in parallel groups) was found to be helpful.

Completeness and applicability

All but four of the studies were conducted in North America. Two^153,158 studies were conducted in the Netherlands, one¹⁶⁵ study in Spain and one¹⁵⁵ study in the UK.

Nine^{155,159–164,166–168} of the seventeen studies focused on sexually abused children, including the one UK study,¹⁵⁵ which compared the effects of two manualised therapies: an individual psychotherapy and a group therapy that included psychoeducation. Six^{151–154,156,158} studies addressed the consequence of witnessing IPV, and two^157,165 studies recruited children who had experienced other forms of maltreatment, including multiple abuse. The studies are broadly relevant to the UK, in terms of participants, settings and transferability of the interventions, but are sometimes limited by the cultural specificity of some interventions (e.g. Holland et al.¹⁶⁴). There is clearly a bias towards evaluating interventions aimed at children who have been sexually abused or exposed to domestic violence, and less evidence about the usefulness of psychoeducational interventions to other groups of maltreated children.

Quality of the evidence

The seven^151–157 trials of psychoeducation are generally of very poor quality, particularly in relation to detection bias and attrition bias (where only the study by Overbeek et al.¹⁵³ was judged low risk of bias), and the quality of the quasi-experiments and COSs is variable. All in all, although the trends generally favour psychoeducation, there is a need for well-designed studies of psychoeducational interventions that are carefully designed in respect of their theories of change, carefully implemented, rigorously evaluated against agreed outcomes and outcome measures, and fully reported.

Study design

One¹⁶⁹ of the five studies was a randomised trial. The remaining were COSs.^170–173

Location of studies

Only one¹⁶⁹ study took place in the UK. Of the COSs, two studies^170,171 took place in Canada and two^172,173 in the USA.

Sample sizes

All studies had small samples sizes. The COSs^170,171 had sample sizes at baseline, ranging from a total of 12 participants¹⁷¹ to 70 participants.¹⁷⁰ Monck et al.¹⁶⁹ included 47 participants.

Participants

Maltreatment

All five^169–173 studies recruited children and young people who had been sexually abused. De Luca et al.¹⁷⁰ and McGain and McKinzey¹⁷² recruited only girls; Grayston and De Luca¹⁷¹ recruited only boys and Monck et al.¹⁶⁹ recruited both boys and girls, although the sample included mainly girls (85%).

Interventions and comparisons

In the De Luca et al.¹⁷⁰ study, participants in the intervention (sexually abused girls) were compared with girls with no known history of sexual abuse. Control group participants in the remaining studies received either TAU, or were placed on a wait-list control.

In four^170–173 studies, the intervention was described as ‘group therapy’. Verleur et al.¹⁷³ assessed the impact of group-based sexual education combined with group psychotherapy (no further information) led by same-sex (female) therapists. McGain and McKinzey¹⁷² set out the goals of the group-based programme, but do not detail the nature of the therapy/group process, other than to observe that it was ‘similar to hundreds of treatment programs provided to children who have been sexually abused’¹⁷² (p. 1168) and to differentiate it from other treatment programmes, such as those that are more behaviourally orientated, brief therapy programmes and crisis intervention programmes. The group intervention for boys in the study by Grayston and De Luca¹⁷¹ was run by therapists of both genders, supervised by a registered clinical psychologist. Sessions were said to follow ‘a consistent four-part format’ (described by De Luca and her associates⁷¹⁷). Behaviour management techniques were used throughout to reduce disruptive behaviour and increase acceptable conduct. At the midpoint, the therapists also implemented a fixed-interval reinforcement schedule to further manage behaviour (described by Hack et al.⁵³⁵).

Monck et al.¹⁶⁹ assessed the effectiveness of a family network treatment programme with or without group treatment.

Number and duration of sessions

Sessions in Grayston et al.¹⁷¹ and De Luca et al.¹⁷⁰ were 90 minutes in duration, and were provided for 12 weeks¹⁷¹ and between 9 and 12 weeks.¹⁷⁰

The group work programme evaluated by McGain and McKinzey¹⁷² continued weekly for 6 months, with the possibility for children to continue on to another cycle. Some children are reported to have spent between 9 months to 1 year in treatment.

The number of sessions ranged considerably, from 9–12 weekly sessions^170,171 to 6–9 months of weekly sessions.^172,173

No information was available on the structure of group work in Monck et al.¹⁶⁹

Depression

Children’s depression was assessed in one¹⁶⁹ study, using the CDI.⁸⁰

Anxiety

Anxiety was assessed in De Luca et al.,¹⁷⁰ using the RCMAS/What I Think and Feel Questionnaire.²⁹³

Behaviour

Problem behaviours were assessed by three^170–172 studies. De Luca et al.¹⁷⁰ and Grayston and De Luca¹⁷¹ both used the CBCL,²⁵⁷ McGain and McKinzey¹⁷² used the Revised Behavior Problem Checklist⁴³⁸ and the ECBI.⁴³⁹

Sexual behaviour

Child sexual behaviour was assessed in Grayston and De Luca¹⁷¹ with the CSBI²⁵⁹ and in Verleur et al.¹⁷³ with the Anatomy/Physiology Sexual awareness scale.

Self-esteem

Children’s self-esteem was assessed by three of the included studies^169,170,173 using the Coopersmith Self-Esteem Inventory (SEI).⁴³⁶

Sequence generation and allocation concealment

Monck et al.¹⁶⁹ stated that ‘allocation was made without the direct involvement of the researchers or the clinical team, and was achieved by blind choice of marked cards drawn from an envelope’¹⁶⁹ (p. 12). Judgements of low risk of bias were made for both sequence generation and allocation concealment.

Blinding of participants, personnel and outcome assessors

This study¹⁶⁹ did not make any reference to procedures to blind the participants or outcomes assessors. However, because of the nature of the intervention, it is unlikely to have been feasible to blind participants. The use of self-report and parent-reported measures also indicates that it was not possible to blind outcome assessors, suggesting this study¹⁶⁹ had a high risk of bias.

Incomplete outcome data

This study¹⁶⁹ analysed only the available data of 47 participants. However, there was insufficient information provided in the study¹⁶⁹ regarding how many participants were originally assigned to each group. Therefore, this study¹⁶⁹ was judged to be of unclear risk of bias.

Selective outcome reporting

There appears to be no evidence that other outcomes were planned and then omitted from the results, but, in the absence of prospective registration, we judged this aspect to be of ‘unclear’ risk of bias.

Other sorces of bias

No other identical sources of bias were identified in these studies.

Summary details of the risk-of-bias assessments of these trials can be found in Appendix 10.

Depression

Monck et al.¹⁶⁹ found no significant difference in depression between children participating in the family network treatment programme with group treatment and those participating in the family network treatment programme alone.

Anxiety

In the only study to assess anxiety, De Luca et al.¹⁷⁰ found that, although anxiety scores for both intervention group (sexually abused) and the comparison group (no sexual abuse) decreased following the intervention, there was no significant difference between the two groups. The authors note that anxiety scores for both groups were low at baseline.

Self-esteem

The results of analyses in De Luca et al.¹⁷⁰ indicated a statistically significant impact of group treatment for self-esteem for those girls who completed therapy. Monck et al.¹⁶⁹ found no difference in measures of self-esteem following the intervention tested in the RCT.

Behavioural problems

Two^170,172 of the three^169,170,172 studies that examined the impact of group therapy on problem behaviour reported post-treatment improvement. Only in De Luca et al.¹⁷⁰ were participants followed up for 1 year. In this study,¹⁷⁰ parents reported sustained improvements in internalising and externalising behaviour problems at 9–12 months’ follow-up, although this reached statistical significance only for internalising behaviours. However, data were not collected on the behaviour of those in the comparison group.

Grayson and De Luca¹⁷¹ also found larger changes in pre–post treatment scores for problem behaviours among children in the intervention group, but the between-group differences were not statistically significant.

Sexual behaviour

Only Grayston and De Luca¹⁷¹ examined the impact of group therapy on sexual behaviour, finding that, although sexualised behaviour tended to decline following treatment, the changes – as measured by the Child Sexual Abuse Inventory – were not statistically different between treatment and control children. Verleur et al.¹⁷³ reported that children in the experimental group improve significantly compared with the control group with regards to the measure of sexual awareness used in the study.

Self-esteem

One¹⁷³ study assessed participants’ self-esteem following the intervention and found that, although the self-esteem for both intervention and control groups increased, there was a larger increase in the treatment group than the control group.

Summary

We identified just five^169–173 studies of group work (all with sexually abused children), of which only one¹⁶⁹ study was a randomised trial. The trial¹⁶⁹ found no evidence of the effectiveness of adding a group work component to a family network treatment in terms of reducing depression or improving children’s self-esteem. For the other studies, the limited information provided about the interventions, combined with heterogeneity in outcomes assessed and measures used, make it impossible to draw conclusions about the effectiveness of therapeutic group work for sexually abused children.

Completeness and applicability

The trial¹⁶⁹ was conducted in the UK. The remaining four^170–173 studies were conducted in North America. All focused on children who have been sexually abused. As indicated above, the studies provide little information that would allow these interventions to be replicated or to explore possible reasons for differences in reported effects.

Quality of the evidence

The trial conducted by Monck et al.¹⁶⁹ ‘scored’ well in terms of risk of bias, and was one of the few studies that reported an adequate sequence generation and concealment of allocation. The COSs were judged to be overall of poor quality.

Study design

Four studies were randomised trials.^{155,174–176} Cadol et al.¹⁷⁷ was a QEx study.

The Downing et al.,¹⁸⁰ Sullivan¹⁷⁹ and Nolan et al.¹⁷⁸ studies were COSs.

Location

Nolan et al.¹⁷⁸ was conducted in Ireland, and Trowell et al.¹⁵⁵ in the UK. The other six studies^{174–177,179,180} were conducted in North America.

Sample sizes

Of the four trials, two were very small. Haight et al.¹⁷⁵ was a feasibility study, with a small sample of just 23 randomised participants (15 completers). Thun et al.¹⁷⁴ recruited 13 participants, but only 11 participants completed the intervention and follow-up assessments. Reddy et al.¹⁷⁶ and Trowell et al.¹⁵⁵ randomised 71 and 75 participants, respectively.

Cadol et al.¹⁷⁷ recruited 140 participants and Sullivan¹⁷⁹ recruited 72 participants. The remaining COSs were very small. Downing et al.¹⁸⁰ had a sample size of 22 participants and Nolan et al.¹⁷⁸ had a sample of 38 participants.

Participants

Maltreatment

Children in the Haight et al.¹⁷⁵ study were in foster care and came from families in which they had experienced multiple forms of abuse, and whose parents misused methamphetamine. In Cadol et al.,¹⁷⁷ the children had experienced both physical abuse and neglect. Downing et al.¹⁸⁰ recruited children who had experienced sexual abuse. Reddy et al.¹⁷⁶ provides no detailed information on maltreatment histories.

Participants in the studies by Trowell et al.,¹⁵⁵ Thun et al.,¹⁷⁴ Sullivan¹⁷⁹ and Nolan et al.¹⁷⁸ had all been subjected to sexual abuse.

Interventions and comparisons

Nolan et al.¹⁷⁸ compared the effectiveness of IT with combined individual and group therapy (IGT). The purpose of these two interventions was to treat the ‘psychological sequelae of child sexual abuse’. Both interventions drew on the principles of a broad range of individual and group therapies (psychodynamic, client centred, CBT, etc.) and implementation varied between therapies and across clients. All therapists sought to provide their clients with a supportive therapeutic relationship that would enable them to process the psychological consequences of sexual abuse, and help them to develop the insights and skills needed to prevent further abuse. Those receiving group therapy also had the opportunity to realise that they were not alone in experiencing CSA, to enjoy peer support, and to benefit from peer-to-peer feedback on their experiences and their behaviour. On average, the duration of intervention ranged from 18 to 20 hours. This was the only intervention received by participants in the comparison group (those in the intervention group also received group psychotherapy).

Sullivan¹⁷⁹ compared group psychotherapy developed at the Boys Town National Research Hospital with individual psychotherapy for survivors of abuse. After listing treatment goals, the reader is referred to an earlier paper for further information.⁷¹⁸ Children in the Sullivan¹⁷⁹ study met weekly with their therapist for 2 hours (because of the communication problems and need for signing) for 36 weeks. Control group participants received no treatment.

Thun et al.¹⁷⁴ state that the group curriculum ‘followed a modified multidimensional model proposed by Lindon and Nourse (1994⁵⁴⁴) that incorporated a skills component, a psychotherapeutic component and an educative component’¹⁷⁴ (p. 8) and, arguably, this study¹⁷⁴ might, with additional information, have been included in the psychoeducational or group treatment grouping. In describing the intervention, Thun et al.¹⁷⁴ emphasise the benefits of groups as a means of alleviating feelings of isolation and alienation, and fostering trust. Those in the comparison group had the option to avail of individual counselling, but none did.

The intervention used by Haight et al.¹⁷⁵ was ‘Life Story Intervention’ (LSI), described by the study authors as a narrative and relationship-based mental health intervention. LSI was delivered in and around the children’s homes, on a one-to-one basis, by a range of professionals, including teachers, child welfare professionals and counsellors. The intervention was delivered over the course of a series of weekly 1-hour sessions for approximately 7 months. Control group participants were placed on a wait-list and received the intervention at the conclusion of the study.

Reddy et al.¹⁷⁶ assessed the impact of Cognitively-Based Compassion Training (CBCT), described by the authors as a type of contemplative practice that is built on mindfulness practice and teaches active contemplation of loving kindness, empathy and compassion towards loved ones, strangers and enemies.⁷¹⁹ It uses a variety of cognitive restructuring and asset-generating practices, with the long-term goal of developing the equanimity of mind that fosters acceptance and understanding of others. Participants were assigned to classes that met twice a week, for 1 hour, for 6 weeks. The control group was a wait-list control.

The interventions used in studies by Cadol et al.¹⁷⁷ and Downing et al.¹⁸⁰ were described as one-to-one counselling. No information was provided on the duration or number of counselling sessions, other than Downing et al.¹⁸⁰ stating that sessions took place ‘near-weekly for approximately 1 year’. In Cadol et al.¹⁷⁷ children in all arms received developmental testing, regular medical care and co-ordination of services, but only those in the experimental arm received the counselling. Participants in the control group in the Downing et al.¹⁸⁰ study received reinforcement treatment (helping parents to focus on positive behaviour).

Trowell et al.¹⁵⁵ compared brief, focused, individual psychoanalytic psychotherapy with group psychotherapy (comprising both psychotherapeutic and psychoeducational components).

Post-traumatic stress disorder

Trowell et al.¹⁵⁵ used the 1982 version of Orvaschel’s PTSD scale – an extension of the KSADS instruments.⁴⁰⁶ Reddy et al.¹⁷⁶ used the Childhood Trauma Questionnaire (CTQ⁴⁴⁸).

Trauma symptoms in Nolan et al.¹⁷⁸ were assessed using the TSCC.²⁶⁰ Specific subscales used included anxiety, depression, anger, post-traumatic stress, dissociation, overt dissociation, fantasy dissociation, sexual concerns, sexual preoccupation and sexual distress.

Depression, anxiety, emotional disorder and conduct disorder

Trowell et al.¹⁵⁵ used a shortened version of the KSADS.²⁶⁸ Reddy et al.¹⁷⁶ used the Quick Inventory of Depressive Symptomatology–Self-Report⁴⁴² to assess depression. Reddy et al.¹⁷⁶ also assessed non-suicidal self- injurious behaviour using the Functional Assessment of Self-Mutilation (FASM⁴⁴³).

Nolan et al.¹⁷⁸ assessed depression using the CDI.²⁵⁸ Specific subscales used included negative mood, interpersonal difficulties, ineffectiveness, anhedonia and negative self-esteem.

Child behaviour

Nolan et al.¹⁷⁸ and Sullivan¹⁷⁹ both assessed problem behaviours using the CBCL.^294,418;720 Specific subscales used in Nolan et al.¹⁷⁸ included total problems, externalising, internalising, withdrawn, somatic complaints, anxious/depressed, social problems, thought problems, attention problems, delinquent behaviour and aggressive behaviour.

Cadol et al.¹⁷⁷ assessed cognitive, physical, social, and emotional functioning of their participants using a range of measures, including the Bayley’s Scales of Infant Development,⁴⁴⁹ Bayley Infant Behavior Record⁴⁵¹ and Child Behavioral Characteristics Questionnaire.

Downing et al.¹⁸⁰ used parents’ and teachers’ behavioural observations to record sleep disturbance, sexual play with other children, enuresis, general misbehaviour and sexual self-stimulation.

Anxiety

Reddy et al.¹⁷⁶ assessed anxiety with the STAI.³³³

Emotional self-regulation

Reddy et al.¹⁷⁶ used the Difficulties with Emotion Regulation Scale (DERS⁴⁴⁶) to assess participants’ awareness and understanding of emotional experience, acceptance of emotions, ability to modulate emotional arousal and effective action in the presence of intense emotions.

Self-efficacy

Reddy et al.¹⁷⁶ used the Children’s Hope Scale⁴⁴⁵ to assess agency and pathways (belief in one’s ability to develop successful call planning).

Self-esteem

Thun et al.¹⁷⁴ assessed participants’ self-image using four subscales (see Results: psychotherapy/counselling) of the Offer Self-Image Questionnaire-Revised (OSIQ-R⁴⁴⁰), a personality test designed to measure self-image of adolescents aged 13–18 years.

Global functioning

Trowell et al.¹⁵⁵ measured social, psychological and school functioning using the KGAS (1986 version) based on the CGAS.⁴⁰⁵

Sequence generation and allocation concealment

Thun et al.¹⁷⁴ stated that participants were ‘randomly assigned’ but gave no other information and Trowell et al.¹⁵⁵ simply said participants were randomised. Both studies were therefore assessed as ‘unclear’ for each domain.

Both Reddy et al.¹⁷⁶ and Haight et al.¹⁷⁵ were assessed as ‘unclear’ risk of bias for each of these domains. Although Reddy et al.¹⁷⁶ state that the study used block randomisation to equalise numbers in each group, no information was provided that shed light on the sequence generation or allocation concealment.

Blinding of participants and personnel

Owing to the nature of the intervention, neither staff nor participants were likely to have been ‘blinded’ in the study by Thun et al.,¹⁷⁴ and the study was therefore assessed as high risk of bias. In Trowell et al.,¹⁵⁵ the authors state it was not possible to blind participants. Haight et al.¹⁷⁵ stated that caregivers were not blind whether or not the child was receiving the intervention, and it would not have been possible to blind the children. Both studies were^155,174 were therefore assessed as high risk of bias. The same judgement was made for Reddy et al.¹⁷⁶

Blinding of outcome assessors

Assessors in Thun et al.¹⁷⁴ were a faculty member and selected graduate students from a Marriage and Family Therapy programme at the University of Southern Mississippi but it is not clear whether or not they were blind to experimental conditions. This item was therefore assessed as unclear.

Reddy et al.¹⁷⁶ provide no information on the blinding of outcome assessors. Haight et al.¹⁷⁵ state that assessments were conducted by master’s level professionals who were not serving as the child’s community clinician, but it is not clear that they were unaware of the allocation status of the children. Both were therefore assessed as ‘unclear’.

Trowell et al.¹⁵⁵ state that blinding of assessors was probably compromised because the children and their mother often mentioned the specific therapy during assesment, so was judged high risk of bias.

Incomplete outcome data

Thun et al.¹⁷⁴ analysed only the available data, as two of the six participants randomised to the treatment arm dropped out. Haight et al.¹⁷⁵ suffered significant attrition and analysed only the available data. Both studies^174,175 were therefore judged as high risk of bias for incomplete outcome reporting.

No information is provided by Reddy et al.¹⁷⁶ other than that caregivers in the wait-list control did not complete, post-treatment, the Inventory of Callous and Unemotional Traits-Parent Report (ICU-P⁴⁴⁷) – one out of six measures of children’s psychosocial functioning. This was therefore assessed as unclear risk of bias.

Trowell et al.¹⁵⁵ was judged low risk of bias on these grounds.

Selective outcome reporting

There appears to be no evidence that other outcomes were planned and then omitted from the results from any of these four^174–176 trials. However, in the absence of a published protocol or trial registration, we have judged all four to be of ‘unclear’ risk of bias.

Other sources of bias

No other potential sources of bias were identified in these studies.

Summary details of the risk-of-bias assessments of these trials can be found in Appendix 10.

Post-traumatic stress disorder

In Trowell et al.,¹⁵⁵ univariate analyses failed to identify a difference between those receiving group and those receiving IT on the PTSD dimension of ‘persistent symptoms of increased arousal’ used in the study⁴⁰⁵ (data not provided, but the authors, following Cohen et al.,⁷¹⁶ used an effect size of 0.5 as a threshold of moderate effect). No further analyses relating to this indicator of PTSD were undertaken.

Controlling for the impairment score on the KGAS⁴⁰⁵ in a multivariate analysis, the authors report a significant effect of IT for ‘the re-experiencing of trauma’ dimension of PTSD (baseline to 1- and 2-year follow up, and baseline to exit), and – for the ‘persistence/avoidance of stimuli’ dimension – at baseline to first-year follow-up. When KGAS is replaced by the baseline score on the same dimensions, the significance of the effects is attenuated.

For trauma symptoms, assessed by Nolan et al.¹⁷⁸ using the TSCC,³²⁵ the only scores to improve were depression (p < 0.05) and anger (p < 0.01), and these improved for participants in both treatments, with no significant difference between the groups. No changes were detected for anxiety, post-traumatic symptoms, dissociation, overt dissociation, dissociation–fantasy, sexual concerns, sexual preoccupation and sexual distress.

Depression

In Nolan et al.,¹⁷⁸ total depression score (p < 0.01), interpersonal problems (p < 0.05) and anhedonia (p < 0.01) all improved for both interventions, but there was no significant difference between groups. ‘Ineffectiveness’ was improved only in the combined IGT group (p < 0.01). Between-group difference in Trowell et al.¹⁵⁵ failed to reach the threshold size of 0.5 adopted by the authors (KGAS).

Behavioural problems

Results in the study by Haight et al.¹⁷⁵ indicate a significant group (experimental or control) by time (pre- or post-test) interaction on child externalising behaviour (p < 0.05), but no main effects. Results indicated that although experimental group externalising scores decreased, control group externalising increased over time. There were no other significant group or time effects for internalising behaviour scores or total problem scores.

Sullivan¹⁷⁹ concluded that participants in the intervention group of this COS had significantly fewer behaviour problems than children not receiving individual psychotherapy. For boys, the treatment main effect was statistically significant for 10 of the 12 dependent variables assessed (total CBL; external and internal composite scales; nine CBL subscale scores). For girls, the main effect was significant for 5 of the 11 relevant variables (total CBL; external and internal composite scales; eight subscale scores). The numbers were very small, particularly for girls, and the authors point to a non-significant main effect for girls for the remaining variables.

Nolan et al.¹⁷⁸ reported that total CBCL scores were reduced for both of the interventions (p < 0.01). In addition, internalising (p < 0.01) and externalising scores (p < 0.05) were also significantly improved by both interventions. CBCL subscales: withdrawn, somatic complaints, anxious/depressed, social problems, attention problems and aggressive behaviour all improved with both interventions (p < 0.01). There was no significant difference between groups. Delinquent behaviour and thought problems remained unchanged. Results from the YSR indicated that there was no significant time or group effect on any YSR scales, indicating that there was no impact of therapy on any of the YSR scales.

Downing et al.¹⁸⁰ reported that for the children in the counselling intervention group, parents reported a decrease in sleep disturbance, sexual play with other children, enuresis and general misbehaviour. However, there was no evident decrease in sexual self-stimulation.

Six weeks after the end of treatment, Reddy et al.¹⁷⁶ found no differences on any measure of psychosocial functioning following CBCT, although the authors suggest that practice frequency was associated with increased hopefulness and trend for decrease in generalised anxiety.

Self-image

Thun et al.¹⁷⁴ found no significant differences between the two groups on the four subscales of the OSIQ-R (Offer et al.⁴⁴⁰) used to measure self-image: impulse control, self-confidence, self-reliance and body image. Negative self-esteem remained unchanged in Nolan et al.¹⁷⁸ (as assessed by the CDI).

For Cadol et al.¹⁷⁷ results of comparisons of the three experimental groups and control indicate significant differences between the groups in the cognitive, physical, social and emotional areas. However, results also indicated that treatment techniques tested do not significantly affect the developmental performance of the participants.

Summary

A relatively small body of evidence was found pertaining to the effectiveness of psychotherapy/counselling interventions. Four^{155,174–176} studies were randomised trials, one¹⁷⁷ study was a QEx study and three^178–180 studies were COSs.

Five^{155,174,178,180,186} of the eight included studies recruited children who had been sexually abused. The remaining three studies recruited children who had experienced a variety of forms of maltreatment. The interventions were disparate, ranging from life story work with children in foster care, to cognitive-based compassion training – intervention built on mindfulness practice and cognitive restructuring.

Most studies compared psychotherapy with supportive counselling or no treatment, including wait-list controls. Nolan et al.¹⁷⁸ compared the effectiveness of IT with combined IGT, and Trowell et al.¹⁵⁵ compared individual psychotherapy with a group therapy that combined psychotherapuetic and psychoeducational components.

One study¹⁷⁶ examined six outcome domains to assess the impact of compassion training, only one of which (depression) was examined in another study,¹⁷⁸ which used a different measure. Four^177–180 studies examined child behaviour as an outcome, but again all used different measures and had very different samples. It is therefore not possible, meaningfully, to draw any overall conclusions about the effectiveness of psychotherapy for maltreated children.

Completeness and applicability

All but Nolan et al.¹⁷⁸ and Trowell et al.¹⁵⁵ were conducted in the USA. Nolan et al.¹⁷⁸ was conducted in Ireland and Trowell et al.¹⁵⁵ in the UK. Most of the interventions described have residents with therapies available in the UK, for example one-to-one counselling,^177,180 the therapies provided in Nolan et al.¹⁷⁸ for sexually abused girls and the LSI described by Haight et al.¹⁷⁵ Given the dominance of counselling and current recommendations to increase access to counselling for troubled children, the dearth of evidence to support the effectiveness of psychotherapy or counselling for maltreated children is of concern.

Quality of the evidence

Four^{174,175,178,180} of the studies, including two^174,175 of the three trials, were extremely small. The trials were poorly reported, with low risk of judgements made only in relation to selective outcome reporting (and this is probably generous). The overall quality of the non-randomised studies was judged to be poor.

Study design

Both^181,182 studies were randomised trials.

Sample sizes

Both^181,182 studies had baseline sample sizes of just 36 and 46 participants, respectively.

Location

Both^181,182 studies took place in the USA.

Participants

Maltreatment

Children in both studies had experienced both physical abuse and neglect.

Interventions and comparisons

The intervention used in both studies^181,182 was resilient peer treatment (RPT), designed to promote the development of social competencies of preschool children in the context of classroom play mediated by a resourceful peer.

In Fantuzzo 1988¹⁸¹ the control group pairs met in the same setting as the treatment group pairs for the same number of play sessions under identical conditions, except that their peer was instructed to respond positively to social initiations but to refrain from initiating social interactions. In Fantuzzo 1996,¹⁸² control group participants were paired with a classmate of average interactive play ability.

Number and duration of sessions

In both^181,182 studies, sessions lasted 15 minutes each, and were spread out over a 2-month period, with no more than three sessions occurring per week.

Outcomes

Interactive peer play was assessed by Fantuzzo 1996¹⁸² using an observational coding system.

Social skills were assessed in the Fantuzzo 1996¹⁸² study, using the Social Skills Rating System (SSRS).³⁷⁹

Problem behaviours were also assessed in Fantuzzo 1996¹⁸² using the SSRS.³⁷⁹

Positive social behaviours were assessed by Fantuzzo 1988¹⁸¹ through the use of the same observational coding system, the SSRS.³⁷⁹

Psychological adjustment was assessed by Fantuzzo 1988¹⁸¹ using the Preschool Behavior Questionnaire.⁴⁵⁵

Pre-academic progress was assessed in Fantuzzo et al.¹⁸¹ using the Brigance Diagnostic Inventory of Early Development.⁴⁵⁶

Sequence generation and allocation concealment

Both trials^181,182 stated their participants were ‘randomly assigned’, but gave no other information on the method of random allocation and allocation concealment. Both were judged ‘unclear’ risk of bias for each of these domains.

Blinding of participants and personnel

One¹⁸¹ study made no reference to procedures to blind participants or personnel, but the other study¹⁸² stated that ‘teachers, play supporters and data collectors were not informed of the maltreatment status of the participants throughout the study’¹⁸² (p. 1284). Fantuzzo et al.¹⁸² was therefore judged to be of low risk of bias. The other¹⁸¹ study was assessed as being of unclear risk of bias.

Blinding of outcome assessors

Both^181,182 studies stated that children were observed by trained raters who were blind to both the assignment of children to conditions and the specific hypotheses of the study, resulting in judgements of low risk of bias.

Incomplete outcome data

Both^181,182 studies were assessed being of low risk of bias, as both were able to minimise potential attrition bias by analysing participants based on ‘assigned treatment’ rather than ‘treatment completed’.

Selective outcome reporting

Both^181,182 studies differed in their choice of outcomes, and there appears to be no evidence that other outcomes were planned and then omitted from the results. However, in the absence of a published protocol or trial registration, we categorised these studies^181,182 as ‘unclear’ risk of bias.

Other sources of bias

No other potential sources of bias were identified in these studies.

Summary details of the risk-of-bias assessments of these trials can be found in Appendix 10.