Cognitive Behavioural Treatment for Anger in Adults with Intellectual Disabilities: A Systematic Review and Meta-analysis

Abstract

Background

The cognitive behavioural treatment for anger in adults with intellectual disabilities has received increasing interest. The current study aims to review the current literature and provide a meta-analysis.

Method

A literature search found 12 studies eligible for the quality appraisal. The studies examined cognitive behavioural treatment for anger in adults with intellectual disabilities published since 1999. Nine studies were eligible to be included in the meta-analysis.

Results

The meta-analysis revealed large uncontrolled effect sizes for the treatment for anger in adults with intellectual disabilities, but is viewed with caution due to low sample sizes. The narrative review showed improved methodological quality of the literature.

Conclusions

The emerging literature is encouraging. However, it is limited through concatenated data, a lack of comparative control groups and small study samples.

Introduction

Cognitive behavioural therapy (CBT) is now the dominant psychotherapy in mental health services, and research has shown the modality to be effective for a wide variety of psychological disorders (Roth & Fonagy 2005). Whilst the general population have benefitted from CBT, people with intellectual disabilities have traditionally had little or no access to psychosocial treatments (Taylor & Lindsay 2007). However, an evidence base for the use of CBT for people with intellectual disabilities is emerging. Prevalence rates of anger-control problems for people with intellectual disabilities are high (Taylor & Novaco 2005), and the treatment for anger problems is becoming one of the most widely researched issues in the field of intellectual disabilities (Willner 2007). This paper will systematically review published reports on the effectiveness of CBT for anger in adults with intellectual disabilities and conduct a meta-analysis.

Anger has been defined as a state of emotion that involves various intensities of feeling, ranging from aggravation and annoyance to rage and fury (Spielberger 1991). It is also recognized that anger is a normal emotion with a particular value, for example increasing motivation (Taylor 2002). The term aggression refers to a range of observable behaviours that are intended to do harm (Berkowitz 1993). The anger construct shares certain properties that overlap with aggression; however, the constructs are not synonymous. The distinction can be summarized by referring to anger as the emotion and aggression as the behaviour (Spielberger et al. 1995).

Amongst people with intellectual disabilities, anger is a frequent problem and often associated with aggression (Willner 2007). More specifically, research has shown that anger is a significant activator of aggression. Whilst anger is reciprocally influenced by aggression, anger is neither necessary nor sufficient for aggression to occur (Novaco 1975; Zillman 1979; Novaco 1994). Novaco (1994) has articulated the relationship between anger and aggression within his cognitive model. He asserts that anger is a subjective emotional state involving cognitions of hostility and physiological arousal and is a causal determinant of aggressive behaviour.

Whilst prevalence rates for aggression are well documented in people who have intellectual disability (Harris 1993; Sigafoos et al. 1994; Smith et al. 1996), anger has rarely been assessed outside of research studies (Taylor 2002). However, Novaco & Taylor (2004) found that anger was predictive of assault in 47% of a male forensic intellectual disability sample (N = 129). The epidemiological and empirical studies reported indicate that aggression and, by implication, anger are significant issues amongst the intellectual disability population.

There is a well-established evidence base for the use of CBT in the general literature (Roth & Fonagy 2005). With regard to the treatment for anger, CBT within the general literature has a growing evidence base reported in numerous meta-analyses (e.g. Beck & Fernandez 1998; Del Vecchio & O'Leary 2004). It has been noted that the treatment of people with an intellectual disability was absent from the ‘What Works for Whom’ literature (Beail 2003). Reviews suggest that research on the efficacy of psychotherapy applied to this population is lacking (Beail 2003; Willner 2005). However, the CBT for anger represents the largest area of interest in terms of published reports in the psychological therapies literature with people who have intellectual disability. Whitaker (2001) reviewed 16 studies conducted between 1978 and 1999, concluding that the evidence for CBT as an effective treatment for anger in people with intellectual disabilities was weak. However, since then, anger treatment has received growing attention in the intellectual disability field, and more recent narrative reviews report an emerging evidence base for the effectiveness of CBT interventions for anger in intellectual disability populations (Willner 2005; Taylor & Lindsay 2007; Willner 2007).

Method

The aim of the current review is to systematically analyse the current literature and provide a meta-analysis of the effectiveness of CBT interventions for anger in adults with intellectual disabilities.

Search strategy

The initial strategy involved the search of three major electronic databases (PsycArticles, MEDLINE and PsycINFO), for the period 1999–2011. Whitaker's (2001) conclusion that the evidence was weak formed the rationale for reviewing studies from 1999 onwards. It was important that the meta-analysis gave an account of the emerging literature, and therefore, the search was restricted from 1999 onwards.

Keywords anywhere in the title for the term ‘anger’ returned 3439 references. To limit the search to the desired population, keywords anywhere in the title for the terms ‘developmental disability’, ‘learning disability’, ‘mental retardation’ and ‘intellectual disability’ returned a total 11 739 references. Each population term was entered individually with the term anger. In conjunction, the four separate searches returned 174 studies.

The 174 studies were screened for content relevance by applying inclusion and exclusion criteria. Studies were included on the following grounds: they (i) were published in English, in peer-reviewed journals; (ii) examined the treatment for anger within a cognitive behavioural framework; (iii) were conducted by qualified staff measuring the dependent variable anger; (iv) were either randomized or non-randomized controlled trials or case series designs, and (v) reported on both male and female participants.

Studies were excluded on the following grounds: (i) interventions that did not use CBT (i.e. any other psychotherapy or pharmacological method); (ii) studies that did not measure anger as the dependent variable; (iii) studies containing populations other than adults with intellectual disabilities (i.e. children with intellectual disabilities; studies from the general literature); and (iv) reviews or other non-primary research.

One hundred and fifty-four studies did not meet the inclusion and therefore were excluded from the study. Therefore, the search strategy yielded 20 relevant studies to be included in the review. The studies that met the inclusion criteria were initially reviewed and rated. However, a closer inspection revealed that some were concatenated. So, where data had been used more than once, duplicate studies were removed. In these cases, the study chosen for inclusion in the quality appraisal was the most recent study, the study with the largest sample size or the study deemed the most reliable.

The 12 studies that met the inclusion criteria and were not concatenated with other studies are listed in Table 1. Data for the review were gathered from the full text copies of the studies and extracted raw data from the studies to calculate effect sizes (ESs) where possible.

Table 1. Cognitive behavioural therapy and anger studies meeting the inclusion criteria and were not concatenated

Author and year	Designa	Setting	Sample/Degree of intellectual disability	Group/Indiv'l	Input	Anger measureb	Outcome	Follow-up	Quality rating (Max 32)
Author and year in bold indicate inclusion into meta-analysis. Author and year in normal type indicate studies excluded from meta-analysis to avoid double counting. ^a RCT, Randomised Control Trial; CT, Controlled Trial; A-B (Pre- and post-intervention – without randomization/comparison). ^b AI, Anger Inventory; PI, Provocation Inventory; IP, Imaginal Provocation; SR, Self Report; CR, Carer Rated; OB, Observed Behaviour; RP, Role Play; PACS, Profile of Anger Coping Skills. ^c Insufficient data to be included in meta-analysis.
King et al. ( 1999 )	Group Study A-B	Community	7 Men 4 Women Mild intellectual disability	Group	32 h	AI (SR) Self-esteem (SR) AI (CR) Behaviour Checklist (CR)	Significant reduction in self-reported anger. Caregivers reported fewer problem behaviours.	Maintained at 12 weeks	20
Howells et al. (2000)c	Case Series	Community	3 Men 2 Women Mild intellectual disability	Group	24 h	Qualitative	No psychometric anger measure. Qualitative evidence suggested all participants more in control of their anger.	None	15
Allen et al. (2001)c	Case Series	Community	5 Women Mild-Borderline intellectual disability	Group	40 h	AI (SR) Recidivism	Significant improvement post-treatment for four participants.	Improved scores maintained at 9 months	15
Willner et al. ( 2002 )	Group Study RCT	Community	9 Women 5 Men Mild intellectual disability	Group	18 h	AI (SR) AI (CR) PI (SR) PI (CR)	Significant improvement within and between groups on self-reported and carer measures of anger reactivity. Improvement associated with verbal IQ and carer attendance at groups.	Continued & significant improvements at 3 months	29
Burns et al. (2003)c	Case Series	Institutional	3 Men 2 Mild intellectual disability 1 Borderline intellectual disability	Group	30 h	AI (SR) AI (SR) Aggression (OB)	Mixed results. All cases showed unstable baselines. Case 1 – mixed evidence of improvement. Case 2 – Mixed evidence of improvement. Case 3 – Some evidence of an increase in scores of anger.	None	19
Lindsay et al. ( 2004 )	Group Study CT	Community	33 Men 14 Women Mild intellectual disability	Group	40 h	AI (SR) Role-Play (OB) Anger Diary (SR) Assault Incidents	Post treatment AI/Diary significantly lower in treatment group compared control. Treatment group showed significant post treatment reductions in role-play scores and fewer assaults.	Scores maintained at 3 and 30 months. No comparison follow-ups	29
Taylor et al. ( 2004 )	Group Study CT	Institution	17 Men Mild – Borderline intellectual disability	Individual	18 h	IP (SR)	IP indices significantly lower in treatment condition compared to waiting list. Scores improved in treatment group.	None	24
Willner et al. ( 2005 )	Group Study CT	Community	Treatment: 5 Men 3 Women Control: Men 7 Women 2 intellectual disability not clear	Group	24 h	PI (SR/CR) (PACS) Coping Skills Inventory (CR)	Significant treatment effects for SR and CR on PI. Significant increase in coping skills for treatment group.	PI (SR/CR) gains maintained at 6 months. Coping skills maintained	29
Taylor et al. ( 2005 )	Group Study CT	Institution	40 Men Mild – Borderline intellectual disability	Individual	18 h	AI (SR) PI (SR) Anger rating (CR)	Some indices of AI and PI in treatment condition declined in comparison to control. Limited evidence from staff re post treatment behaviour.	Gains maintained on selected indices at 4 months	29
Hagiliassis et al. ( 2005 )	Group Study RCT	Community	Treat'nt: n = 14 7 Men 7 Women Comparison: n = 15 9 Men 6 Women intellectual disability Wide Range	Group	32 h	AI (SR)	Significant pre- and post-change on anger measure v comparison group. Significant pre- and post-treatment change.	Gains maintained at 4 months	28
Rose et al. ( 2005 )	Group Study CT	Community	Treat'nt: n = 50 40 Men 10 Women Control: n = 36 31 Men 5 Women Level of intellectual disability <70	Group	32 h	AI (SR)	A significant pre- and post-treatment effect between groups. Reliable change showing improvement from 11 participants.	None	23
Rose et al. ( 2008 )	Group Study CT	Community	Treat'nt: n = 20 13 Men 7Female Control: n = 21 16 Men 5 Women Level of intellectual disability <70	Individual	Variable (up to 18 h)	AI (SR)	Significantly reduced anger pre/post and in comparison with control. Reliable change was seen in 7 of 20 participants.	None	25

Data analysis

Two forms of analyses were carried out: (i) quality appraisal; and (ii) a meta-analysis of ESs.

Quality appraisal

A checklist by Cahill et al. (2010), adapted from Downs & Black (1998), was employed to assess the methodological quality of the studies. The checklist was originally designed for the appraisal of both randomized and non-randomized studies and subsequently adapted to make it more relevant to practice-based research. The current review used the adapted version to allow for an assessment of a wider range of methodological designs.

The checklist provides an overall quality score (maximum 32) and four separate subscales:(i) Reporting: this subscale assesses the extent to which the reader is able to make an unbiased appraisal of the findings; (ii) External Validity: a subscale to consider whether the findings can be generalized; (iii) Internal Reliability: a subscale to consider the rigour of the measurement of the intervention; (iv) Internal Reliability (Sampling): a subscale assessing confounding variables including sample bias.

The first author rated each of the studies a graduate psychologist acted as an independent rater, rating a random subsample of four studies. Pairwise agreement calculated K = 0.63. A Kappa rating within this range is considered ‘good’ within the classification of agreement levels indicated by Landis & Koch (1977).

Meta-analysis

Whilst it is acknowledged that uncontrolled ESs are deemed to have less rigour than controlled ESs, a pragmatic decision was made to use uncontrolled ESs. The rationale for this was threefold. Firstly, by analysing only treatment arms of studies, the review would broaden its reach to include as many studies as possible (e.g. case series designs). This was important, as the intellectual disability literature is relatively small in comparison with the general literature. Secondly, it was noted that the control groups across studies differed. Whilst some studies in the literature utilize treatment as usual control groups (Taylor et al. 2004), other studies employ waiting list controls (Rose et al. 2008), whilst others use work group controls (Willner et al. 2005). It was decided that these non-equivalent control groups were not comparable. Finally, data for the calculation of controlled ESs were not always available. Overall, it was decided that comparing pre- and post-treatment data, and reporting uncontrolled ESs, would facilitate a more meaningful review of the evidence.

Inspection of the literature revealed that some studies used multiple outcome measures. Where this was the case, a decision was made to use either the primary outcome measure in the study, or otherwise, the most commonly used and validated measure.

Recent narrative reviews (Taylor & Lindsay 2007; Willner 2007) have reported emerging evidence for cognitive behavioural treatment for anger in intellectual disability populations. It is of note, however, that the studies in the intellectual disability literature report on relatively small sample sizes, particularly in comparison with the general literature (see Roth & Fonagy 2005). The rationale for conducting a meta-analysis was to attempt to increase the statistical power of the emerging literature by quantifying an overall ES of treatment. However, Cohn & Becker (2003) caution that individual ESs calculated from small sample sizes are limited by large confidence intervals (CIs) and therefore less precision. For this reason, they are viewed more tentatively than meta-analyses with large sample sizes that are able to provide more accurate estimates of treatment effect. This guidance is acknowledged in the current report in view of the small Ns of the studies included in the meta-analysis. To some extent, however, this limitation is countered by the meta-analytic method employed by the current study.

stata software (Stata 10, Stata Corp. 2001) is a general-purpose statistical package that includes a number of meta-analytic methods (Egger et al. 2003). The ‘metan’ command produces ESs and their 95% confidence interval (CI) for each study and presents them as a forest plot. ESs were calculated by dividing the difference in mean values pre- and post-therapy by the pre-treatment standard deviation. The influence of each study's ES on the overall ES is weighted according to the sample size. The programme also produces a statistic to assess the heterogeneity of the included studies (Egger et al. 2003).

A fixed-effect meta-analysis was used. This model was chosen because it makes the assumption that the true effect of treatment is the same value in each study (i.e. the effectiveness of CBT is fixed). Therefore, any differences between studies would result purely from the play of chance (Egger et al. 2003). Cohn & Becker (2003) illustrate how fixed-effect meta-analyses increase statistical power by reducing the standard error of the weighted average ES. This is advantageous for the current analysis that reports on studies with small Ns. All ESs were gathered from the published figures provided in the studies.

Benchmarks

Ideally, a treatment benchmark would be an ES that had been established for an emotional problem (e.g. depression) within the intellectual disability literature. However, there is no such literature within the intellectual disability field. The review will turn to the general literature where there have been a number of studies examining CBT treatment for anger problems. In offender populations, De Guiseppe & Tafrate (2003) revealed large controlled ESs (0.73 and 0.99) in their meta-analysis of 57 studies across 1841 adult participants (average ES = 0.86). Del Vecchio & O'Leary (2004) reviewed 23 studies involving 1340 non-institutionalized adults receiving CBT for anger. The analysis revealed mean weighted (large) controlled ES across the studies ranging from 0.61 to 0.90 (average ES = 0.76).

Results

After the removal of the duplicated studies, 12 studies were included in the quality appraisal (10 studies were conducted in the UK, and two were conducted in Australia). Table 1 shows summary information and quality ratings from the studies. Three studies did not provide sufficient data to be entered into the meta-analysis and are indicated in Table 1. This left nine studies that were entered into the meta-analysis, highlighted in bold in Table 1. The review will firstly present the results from the quality appraisal before presenting the outcomes in the meta-analysis.

Quality appraisal of studies

The methodological rigour of each study (N = 12) was appraised using the Cahill et al. (2010) criteria. Table 2 shows the combined scores of the individual studies (also expressed as percentages). The percentage quality scores and themes that emerged within each subscale of the checklist are then reported and provide a context to the meta-analysis.

Table 2. Scores and corresponding percentages of checklist criteria addressed by the studies

Quality criteria subscale
Studies	Reporting score, %	External validity score, %	Internal reliability score, %	Internal reliability sampling score, %	Overall score, %
All studies (N = 12)	109/132, 82.5	83/132, 62.8	50/60, 83.3	42/60, 70	284/384, 74.6
Studies inc. meta-analysis (N = 9)	87/99, 87.8	65/99, 65.6	42/45, 93.3	41/45, 91.1	235/288, 84.5
Group (N = 6)	58/66, 87.8	44/66, 66.6	29/30, 96.6	26/30, 86.6	157/194, 84.4
Individual (N = 3)	29/33, 87.8	21/33, 69.6	13/15, 86.6	15/15, 100	78/96, 86

The overall level of quality criteria addressed by all studies was 74.6%. The studies included in the meta-analysis had a higher overall quality rating of 84.5%. When the studies were split into the two separate meta-analyses (group and individual), the overall quality criteria addressed were 84.4 and 86%, respectively. Individual ratings are indicated in the final column of Table 1. Out of a possible maximum score of 32 points for each individual study, the mean quality rating score across all studies was 23.75 (SD = 5.39).

Reporting

The reporting subscale assessed whether studies provided sufficient information to make an unbiased decision regarding the findings from the review. The overall quality rating addressed from all the studies on this scale was 82.5%, making the reporting subscale a relative strength of the literature. In general, the studies provided clear hypotheses and aims, with clarity over the dependent variable (i.e. anger), and main findings are clearly presented. The characteristics of the client group are well illustrated, and the CBT interventions are, on the whole, clearly described with appropriate references for further details.

All studies reported utilizing a cognitive behavioural framework based on the work of Novaco (1975, 1994). Studies were given higher scores if a detailed account of the intervention package was provided. In particular, the studies included in the meta-analysis revealed treatment packages that included the following CBT strategies: psycho-education, self-monitoring, cognitive restructuring, relaxation, self-instruction, role-play and problem solving. Stress inoculation, assertiveness training, emotional recognition and self-instruction were evident but to a lesser extent.

As CBT is presented as a package, it is not clear what part of the package exerts the biggest impact on change. The active ingredients of therapy have been discussed within the literature (Sturmey 2004). The debate surrounds whether people with an intellectual disability benefit from cognitive therapy over and above that of behaviour therapy (e.g. relaxation). It is beyond the scope of the current review to debate this fully, and the studies reviewed do not generally attempt to decipher this.

Evidence is emerging, however, of variables that may mediate the change process. Willner et al. (2005) reported that their instrument [Profile of Anger Coping Skills (PACS)] revealed relaxation (behavioural intervention) was only utilized by one participant. Whilst it was acknowledged that this was a shortcoming of the intervention, it also allowed for an interpretation of the intervention that did not include relaxation. The authors hypothesized that the ingredients of change were cognitive restructuring, assertiveness training and distraction techniques. Willner et al. (2002), along with Willner et al. (2005) and King et al. (1999), report the benefits of carer participation and comment that this can help clients apply lessons learned during the intervention process, to real life settings.

Overall, the results from the reporting subscale indicate that the studies are well described, and the interventions are clearly presented. It is apparent that whilst there is debate regarding the active ingredients of therapy, there is evidence identifying aspects of interventions that are having an impact on the change process.

External validity

The external validity subscale concerns whether studies can be generalized. The overall quality rating addressed from all the studies on this scale was 62.8%. A variable of interest is the level of ability of participants included in the studies. Eight studies report their samples to be within the mild level of intellectual disability. Hagiliassis et al. (2005) include in their sample participants with both moderate and severe intellectual disabilities.

The relationship between IQ and treatment effectiveness is becoming a specific focus within the literature. Rose et al. (2005) assessed verbal IQ (VIQ) as a specific variable, finding a significant correlation between higher VIQ reported and better treatment outcome. Whilst Rose et al. (2005) and Willner et al. (2002) showed a significant correlation, Willner et al. (2005) also showed a link but not a statistically significant relationship.

Overall, the external validity subscale illuminates the importance of identifying the level of ability of participants when generalizing the findings. The results from this section indicate the effectiveness for anger treatment in adults (in particular) with mild intellectual disabilities. Furthermore, there is emerging evidence that anger treatment may be effective for adults with moderate and severe intellectual disabilities (Hagiliassis et al. 2005). There is also emerging evidence of a correlation between VIQ and treatment outcome.

Internal reliability

The overall quality rating addressed from all the studies on this scale was 83.3%. Taylor (2002) concluded there was a lack of valid and reliable measures for diagnosing anger problems. However, an examination of the internal reliability subscale showed that 10 of the 12 studies included in the quality review utilized well-validated psychometric tools, for example the Novaco Anger Scale (NAS; Novaco 2003) and the Benson Anger Inventory (Benson & Ivins 1992).

The main outcome measures are self-report. An issue here is the limitation of self-reporting in that scoring can be easily minimized or exaggerated by the participant. There is concern, especially within forensic populations, that participant responses may be vulnerable to bias due to participant perceptions that being truthful about anger may hinder their progress through the care pathway. Taylor & Novaco (2005) discuss the parameters within which interviews are optimally conducted to help minimize bias. The current review rated studies accordingly, mindful of the care given by study authors to account for this potential confounding variable.

Examination of the literature revealed the emerging diversity in tools used by researchers to measure anger. Reference to Table 1 shows that studies have supplemented self-report instruments by including carer rated measurement tools (e.g. Willner et al. 2005), observer-rated behaviour of role-play situations (e.g. Lindsay et al. 2004) and imaginal provocation tests (Taylor et al. 2004).

Overall, examination of the external validity subscale shows there has been a concerted effort to adapt and create valid and reliable measurements for the assessment of anger within the intellectual disability population. This has been particularly successful with self-report measures. It is common within the literature for studies to report the validity and reliability of measures. The quality of the assessment tools to measure anger has become a relative strength in the literature.

Internal reliability sampling

The internal reliability sampling subscale addressed bias in the selection of participants. The quality rating addressed from all the studies on this scale was 70.0%. Analysis of this subscale illuminates the design method employed by the studies. Whitaker (2001) and Taylor (2002) noted that the literature lacked sound methodological designs. There had been a call for studies to aim for the gold standard of the randomized controlled trial (RCT).

Eleven controlled studies made up the original 20 studies that met the inclusion criteria. However, as discussed, there is some ambiguity here, as many of the studies are concatenated. After the removal of duplicate studies, eight control studies have un-concatenated data. All studies report the treatment to be effective. However, randomization was only attempted in two studies (Willner et al. 2002; and Hagiliassis et al. 2005). The remaining studies are therefore open to critique as the participants receiving the two treatments are likely to be systematically different. An attempt to use RCTs in the literature is positive; however, the RCT of Willner et al. (2002) reports on a small sample size (N = 14) and is viewed with some caution for this reason. Furthermore, observer ratings were used, and it is evident that the researchers were not blind to the treatment; thus, the research is open to the opportunity for personal bias to intrude.

The RCT of Hagiliassis et al. (2005) offers the most reliable evidence to suggest the effectiveness of anger treatment in their group study of 29 people with a range of intellectual disabilities. They report significant treatment effects that are maintained at 4-month follow-up. The study, however, may have benefitted from a second placebo comparison group to demonstrate that improvement was not just an artefact of ‘attending a group’.

The eight control studies utilized control groups drawn from waiting lists, treatment as usual, routine care, as well as work groups. The impact of these control groups is unclear. This is particularly evident in the studies conducted in institutionalized settings where participants are living together (Taylor et al. 2004, 2005). It is questionable whether the effects of treatment are contained within the experimental group, as participants are in regular contact with each other between therapy sessions. An inspection of Taylor et al. (2005) reveals unexplained treatment gains in the control group.

Taylor and his colleagues have augmented the literature in forensic settings. Here, a rationale for treatment is to reduce anger levels that would in turn reduce aggression and recidivism. A limitation of studies on offender populations is the lack of a comparison group of people with intellectual disabilities who have not offended (Lindsay 2002). Subsequently, whilst these studies show the effectiveness of anger treatment, it is uncertain whether the treatment would meet the criteria for reliable and clinically significant improvement (Jacobson & Truax 1991). That is, whether treatment with this population would reduce anger levels, moving them from a clinical (offending) range, to a non-clinical (non-offending) range.

Overall, the literature has responded to early narrative reviews that called for more methodologically sound designs. However, the evidence is diluted by the amount of concatenated data. In regard to offender populations, there is a need for studies to include comparison control groups of non-offenders. Finally, the lack of conformity of control groups between studies makes it problematic when comparing one study with another. This concern was a decisive factor when employing uncontrolled ESs in the meta-analysis.

Outcomes: effect size meta-analysis

From a total of 20 studies that met the inclusion criteria, 17 studies provided data to calculate ESs. To avoid double counting, eight studies from the 17 were excluded. This left nine studies to be included in the meta-analysis (six studies reported on group-based treatment; three studies reported on individualized treatment). The analysis will be presented in three parts: (i) analysis of all nine studies; (ii) analysis of the six group-based studies; (iii) analysis of the three individual treatment studies.

The outputs produced by stata (Stata 10, Stata Corp. 2001) presented in Figures two, three and four show the name of each study in the left column. For each study, the ES, corresponding 95% CI and the percentage contribution to the overall sample size are listed. Below, the list of studies is the aggregated ES and its 95% CI (illustrated by the diamond shape). Calculated from the means and standard errors of each study, the I-squared percentage indicates the percentage of the overall ES that may be due to heterogeneity between the studies. The P-value is an indicator of the statistical significance of the heterogeneity. Inspection of the P-values of all three meta-analyses shows that the heterogeneity of the included study samples is not significant (i.e. the study samples are similar, homogenous).

All studies

The first part of the analysis included all nine studies. The average ES [overall standard mean difference (SMD)] across these studies (k = 9) was 0.88 (95% CI = 0.65–1.12, N = 168) and ranged from SMD = 0.39 (95% CI = −0.61–1.39) to SMD = 1.29 (95% CI = 0.55–2.03); I squared = 0.0%, P = 0.922.

It is of interest to note the large CIs reported amongst the studies. An interpretation of the large CIs is given here but is relevant for all three parts of the meta-analysis. Figure 1, for example, shows that Willner et al.'s study (2002) with a sample size of eight has a confidence interval that ranges between −0.58 (a negative treatment effect) and 1.59 (a large treatment effect). This study, however, was small and represented only 4.72% weight contribution to the overall meta-analysis. In contrast, studies by Lindsay et al. (2004) and Rose et al. (2005) are larger studies with narrower 95% CIs of 0.45–1.53 and 0.40–1.25, respectively. Consequently, these studies have a greater influence on the overall ES.

Figure 1. Group studies meta-analysis results.

Group studies

In total, six studies were entered into the group studies analysis (see Figure 1). The average ES across the studies (k = 6) was 0.84 (95% CI = 0.57–1.12, N = 123) and ranged from SMD = 0.39 (95% CI = −0.61–1.39) to SMD = 1.01 (95% CI = 0.07–1.96); I squared = 0.0%, P = 0.899.

Individual studies

In total, three studies were entered into this part of analysis (see Figure 2). The average ES across the studies (k = 3) was 1.01 (95% CI = 0.54–1.48, N = 45) and ranged from SMD = 0.70 (95% CI = −0.03–1.44) to SMD = 1.29 (95% CI = 0.55–2.03); I squared = 0.0%, P = 0.545.

Figure 2. Individual studies meta-analysis results (N = 3).

Discussion

The main aim of the review was to systematically evaluate the quality of the literature and provide a meta-analysis of cognitive behavioural anger interventions for people with intellectual disabilities. The number of studies that were initially identified for the review was relatively small (N = 20). Inspection of the studies revealed that eight studies had to be removed due to concatenated data. Therefore, 12 studies were identified for the quality appraisal. Of these, nine provided sufficient data to be included in the meta-analysis.

Rated quality of the data

The studies included in the review were characterized by high levels of reporting and high levels of internal reliability in accordance with the Cahill et al.'s checklist (2010). Examination of these two subscales revealed that the studies included reported sufficient information for the reader to make unbiased assessments of their findings. The high internal reliability score reflects the improvement in the integrity of the tools now being utilized to measure the anger construct. This was reportedly lacking in early narrative reviews (Whitaker 2001).

A relatively lower overall score was returned on the internal reliability sampling subscale. This low score is partly an artefact of the unresponsiveness of the checklist to studies that do not include a randomized control group. Here, a significant proportion of the studies (almost 34%) were marked down because they did not employ a control group; almost 84% did not utilize a randomized control group. The external validity subscale returned a lower score partly due to the problem of generalizing results from studies, which are based on participants with specific ability levels, to the wider intellectual disability field.

A close examination of the checklist did, however, prove useful in illuminating the common themes reported in the results section. It is also of note that the studies included in the review did not provide information about whether study participants had been treated with medication. It may well be that this may be an important factor in the treatment for anger. However, research has shown that there is little evidence to suggest that medication can reduce levels of anger and aggression (Matson et al., 2003). Overall, whilst the studies had certain idiosyncrasies, they also shared important similarities and common principles, which enabled nine studies to be included in the meta-analysis.

Meta-analysis

Of the 12 studies that were included in the quality appraisal, nine provided sufficient data to be considered for the meta-analysis. This is in stark contrast to the array of cognitive behavioural studies for emotional problems in the general population (Roth & Fonagy 2005). In consideration of the emerging anger literature, it is of upmost importance to clinicians working in the intellectual disability field to have an understanding of the evidence for the effectiveness of CBT interventions for anger problems. The current review yielded an overall uncontrolled ES of 0.88. Analysis of group treatment alone yielded an uncontrolled ES of 0.84, and analysis of individual treatment formats yielded an uncontrolled ES of 1.01. These are considered to be large ESs (Cohen, 1992).

Before a comparison is made to benchmark studies, it is important to discuss the nuances of the analysis that inform a more measured interpretation of the large ESs reported. It is of note that the meta-analysis has calculated ESs for studies with small sample sizes (i.e. the total N for the overall analysis is 168). Inspection of the results reveals that individual studies in the meta-analysis (e.g. Willner et al. 2002; Willner 2005; Taylor et al. 2005) had very small samples, resulting in wide 95% CIs (Cohn & Becker 2003). Individually, the small studies produce ESs, which are less precise than larger studies. Indeed, their 95% CIs indicate that there may be no positive treatment effect.

However, by aggregating the individual study sample sizes, the meta-analysis produces a larger sample with which to assess the overall ES and 95% CIs (Egger et al. 2003). As there was no significant heterogeneity between the studies, this aggregating would seem acceptable.

Even so, whilst the overall ES for all of the studies is 0.88, the lower confidence interval is 0.65. In view of the limitations of the sample sizes, it may be prudent to caution towards the 0.65 ES as a more conservative representation of treatment effect. Similarly, it may be prudent to caution towards the lower confidence intervals for the group and the individual analysis, which were 0.57 and 0.54, respectively.

In view of this cautious approach, the ESs are closer to medium in size (Cohen, 1992) and are modest in relation to benchmark equivalent studies (De Guiseppe & Tafrate 2003; Del Vecchio & O'Leary 2004), which returned controlled ESs of 0.86 (N = 1841) and 0.76 (N = 1340) in their large studies. A reflection here is that the anger literature in the current review is restricted by the small study samples. Furthermore, restricted sample sizes are likely to be an ongoing obstacle when conducting research in the intellectual disability field due to the relatively small population.

Whilst it provided a useful template for the analysis of the studies, the adapted quality review checklist (Cahill et al. 2010) was found to be unresponsive to some of the nuances of the data provided in studies that were not controlled. With regard to the meta-analysis, Weston & Morrison (2001) argue that uncontrolled ESs, of the kind reported in the current review, confound effects of the passage of time, effects of common factors and genuine placebo effects. The impact of this is that the current findings are viewed more tentatively with this limitation in mind. As discussed in detail, the analysis was also limited due to the small sample sizes in the studies included in the meta-analysis. Furthermore, the systematic review was constrained by the concatenated data within the literature whereby nine studies met the criteria for inclusion into the meta-analysis after a further eight studies were removed to avoid double counting. Despite these issues and limitations, in consideration of whether the results support the effectiveness of cognitive behavioural interventions for anger problems in adults with intellectual disabilities, the findings indicate at least a medium treatment effect.

Conclusion

The results from the review reveal an emerging evidence base for cognitive behavioural anger interventions in adults with intellectual disabilities. The quality appraisal revealed that studies are now utilizing reliable and valid measurements of the anger construct that were highlighted as missing in previous reviews (Whitaker 2001). Furthermore, the quality appraisal revealed a good level of methodological rigour in the studies, especially the studies that were then entered into the meta-analysis. The literature is, however, limited by concatenated data, which in turn limited the inclusion of studies into the meta-analysis.

The results from the meta-analysis revealed large ESs for the cognitive behavioural treatment for anger in adults with intellectual disabilities. The results of the meta-analysis, however, are viewed with caution. Firstly, the results are limited due to the small number of participants recruited in the studies analysed. This has an effect of creating large ES variability amongst the studies. Whilst the meta-analysis makes an adjustment for this variability, it is advised that the ESs are interpreted cautiously and that particular attention is given to the lower ranges of the ESs reported. Furthermore, the literature is still not at a point where there is a commonality between studies that allows for the calculation of controlled ESs. Thus, the current analysis was only able to provide uncontrolled ES that is deemed less robust than their controlled ES equivalents.

The meta-analysis results illuminated the limitations of research conducted on relatively small sample sizes. It would be helpful for future studies to be designed with larger sample sizes to strengthen assertions made about the effectiveness of anger treatment. Furthermore, a limitation identified by the review is the concatenated data in the literature. Whilst it is acknowledged that the intellectual disability population is relatively small, attempts to overcome this limitation with data from new samples of participants would strengthen the evidence base. It would also be useful for researchers to agree on one type of control group in the research that would allow for a meta-analysis of controlled studies. Moreover, it would be useful for researchers to consider study designs that examine the individual components of treatment packages to further understand the active ingredients of therapy.

Correspondence

Any correspondence should be directed to Matthew Nicoll, Clinical Psychologist, Men's Personality Disorder Service, Nottinghamshire Healthcare, Trust William Tuke House, Rampton Hospital, Retford, Nottinghamshire DN22 OPD, UK (e-mail: [e-mail]).