Non‐pharmacological interventions for challenging behaviours of adults with intellectual disabilities: A meta‐analysis

Abstract Background Non‐pharmacological interventions are recommended for the treatment of challenging behaviours in individuals with intellectual disabilities by clinical guidelines. However, evidence for their effectiveness is ambiguous. The aim of the current meta‐analysis is to update the existing evidence, to investigate long‐term outcome, and to examine whether intervention type, delivery mode, and study design were associated with differences in effectiveness. Method An electronic search was conducted using the databases Medline, Eric, PsychINFO and Cinahl. Studies with experimental or quasi‐experimental designs were included. We performed an overall random‐effect meta‐analysis and subgroup analyses. Results We found a significant moderate overall effect of non‐pharmacological interventions on challenging behaviours (d = 0.573, 95% CI [0.352–0.795]), and this effect appears to be longlasting. Interventions combining mindfulness and behavioural techniques showed to be more effective than other interventions. However, this result should be interpreted with care due to possible overestimation of the subgroup analysis. No differences in effectiveness were found across assessment times, delivery modes or study designs. Conclusions Non‐pharmacological interventions appear to be moderately effective on the short and long term in reducing challenging behaviours in adults with intellectual disabilities.


Introduction
Non-pharmacological interventions for challenging behaviours of adults with intellectual disabilities are being recommended as first line treatments by several leading clinical guidelines (Banks & Bush, 2016; National Institute for Health and Care Excellence, 2017; Sullivan et al., 2018). Moreover, health care professionals prefer non-pharmacological interventions to pharmacological treatments for the management of challenging behaviours (Unwin & Deb, 2008). However, the evidence on the effectiveness of non-pharmacological interventions for challenging behaviours of adults with intellectual disabilities remains unclear. In the past decades, much of the intervention research focused on children and adolescents rather than on adults (Brosnan & Healy, 2011;Heyvaert, Meas, & Onghena, 2010;McIntyre, Blacher, & Baker, 2002), and concerned studies that lacked follow-up measures (Brosnan & Healy, 2011;Chan et al., 2010), with small sample sizes (Didden, Korzillus, van Oorsouw, & Sturmey, 2006;Hassiotis & Hall, 2008;Heyvaert, Maes, van den Noortgate, Kuppens, & Ongehena, 2012) and uncontrolled designs (Allen & Tynan, 2000). Only recently, studies with larger adult sample sizes and (randomised) control groups have been published (Hassiotis et al., 2018;MacDonald, McGill, & Murphey, 2018;McGill et al., 2018;Singh et al., 2018). These studies have not yet been included in the most recent meta-analysis (Knotter et al., 2018), which found that staff training does not reduce challenging behaviours of individuals with intellectual disabilities. Combining early and more recent findings is warranted, in order to gain reliable and up to date insight into the effectiveness of non-pharmacological interventions.
Approximately 10-20% of adults with intellectual disabilities show challenging behaviours (Emerson et al. 2001;Bowring, Totsika, Hastings, Toogood, & Griffith, 2017), including aggression, disruptive and socially inappropriate behaviours, self-injury and withdrawal behaviours (Hartley & MacLean, 2007;Lundqvist, 2013). They can be long-lasting and harmful for the quality of life of the individual concerned (Cooper et al., 2009;Heyvaert et al., 2010). Individuals with intellectual disabilities and challenging behaviours are at higher risk of abuse, neglect, deprivation, institutionalisation, and physical and chemical restraints, compared to individuals with intellectual disabilities without challenging behaviours (Sturmey, 1999;Emerson et al., 2001;Robertson et al., 2005;Holden & Gitlesen, 2004). Besides, challenging behaviours may negatively affect the immediate environment of the individual concerned. Caregivers may be subjected to verbal and physical abuse, or to witnessing self-injurious behaviours (Lambrechts & Maes, 2009). These experiences may cause anxiety, anger, fear and emotional exhaustion (Allen & Tynan, 2000;Smyth, Healy, & Lydon, 2015;Strand, Benzei, & Saveman, 2004). Additionally, staff working with individuals with intellectual disabilities and challenging behaviours report to feel impaired in providing sufficient care (Hartley & MacLean, 2007). Therefore, it is of the utmost importance to treat these behaviours.
The evidence for the effectiveness of non-pharmacological interventions to reduce challenging behaviours in adults with intellectual disabilties is ambiguous. Whereas some previous reviews and meta-analyses found that non-pharmacological interventions are indeed effective in reducing challenging behaviours (Brosnan & Healy, 2011;Didden et al., 2006;Harvey, Boer, Meyer, & Evans, 2009;Heyvaert et al., 2010;Heyvaert et al., 2012;Shogren, Faggella-Luby, Jik, & Wehmeyer, 2004), others did not (Gustafsson et al., 2009;Hassiotis & Hall, 2008;Chan et al., 2010;Cox, Dube, & Temple, 2015;Knotter et al., 2018). These contradictory findings may be due to the scarcity of high quality studies included in previous reviews and meta-analyses. Another explanation might be the heterogeneity in non-pharmacological interventions, as these include various treatments of different theoretical backgrounds. Examples include treatments directed at the individual such as multisensory therapy, mindfulness or cognitive behavioural therapy (CBT; Hassiotis & Hall, 2008;Lotan & Gold, 2009;Chan et al., 2010;Hwang & Kearney, 2013;Nicoll, Beail, & Saxon, 2013), and interventions directed at the environment, such as staff training, applied behaviour analysis (ABA), positive behaviour support or specialised teams (Hassiotis et al., 2009;Knotter et al., 2018;LaVigna & Willis, 2012;MacDonald & McGill, 2013). Moreover, some non-pharmacological interventions are adapted to the specific individual and his or her context, usually by means of a functional analysis of the behaviour of the individual (e.g. ABA or positive behaviour support), while others are more generic programs (e.g. multisensory therapy). Recent studies found positive effects of environmentally mediated positive behaviour support with or without mindfulness components (MacDonald et al., 2018;McGill et al., 2018;Singh et al., 2018). These studies were published after the most recent meta-analysis, which found that staff training has no effect on challenging behaviours of adults with challenging behaviours (Knotter et al., 2018).
The current study was primarily aimed at updating the existing evidence on the effectiveness of non-pharmacological interventions to treat challenging behaviours in adults with intellectual disabilities. Secondary aims were to investigate long term treatment effects, and to examine whether intervention type (i.e. interventions of different theoretical backgrounds) and delivery mode (i.e. individual interventions or environment mediated interventions) were associated with differences in treatment effects. Furthermore, we aimed to investigate whether study design (i.e. randomised versus non-randomised) was related to differences in outcome.

Registration and literature search
The current meta-analysis was registered at PROSPERO (registration number: CRD42016051263; https://www.crd.york.ac.uk/ PROSPERO/display_record.php?RecordID=51263). We included studies that 1) reported on the evaluation of one or more non-pharmacological intervention(s), primarily aimed at reducing or eliminating challenging behaviours of adults with intellectual disabilities (regardless of other diagnoses); 2) included a sample with at least 75% of participants of 18 years or older; 3) used an experimental design (randomised controlled trial; RCT) or quasi-experimental design (pretest-posttest or controlled study) with at least 15 participants; 4) were English-written; 5) were published in an academic, peer-reviewed journal; 6) contained sufficient data to perform meta-analyses (i.e. pre and posttest means, standard deviations, sample sizes, and odds ratios and/or correlations).
In order to be able to investigate a rather homogeneous sample of adults with intellectual disabilities and challenging behaviours, we excluded studies in forensic settings or with forensic participants. Delinquent adults with intellectual disabilities differ in aggression levels compared to non-delinquent adults with intellectual disabilities . By excluding the forensic population, our results would be more specifically applicable to the general care for adults with intellectual disabilities and challenging behaviours.
We used the EBSCOHOST databases Medline, Eric, PsychINFO and Cinahl and searched echt electronic database separately, after which duplicates were removed. Furthermore, reference lists of relevant systematic reviews and meta-analyses were hand-searched to check for possible missing articles. We completed the search on November 14 th 2019. Table 1 displays the search terms used for the databases. Only two limits were applied: publication type (academic journals only) and the publication language (English).
The first author (EB, PhD student) screened all search results on their eligibility in a three-step process: screening based on the title, based on the abstract, and based on the full text paper. The excluded articles were checked by the last author (AdB, senior researcher) and disagreement was resolved through consensus. If agreement could not be achieved, the second author (BJvdH, professor) was consulted. Data extraction was done by the first author. In case data were insufficiently described in the paper, authors were contacted by e-mail or through Researchgate (https://www.researchgate.net/). The following study characteristics were recorded from the included studies: 1) participant characteristics (level of intellectual disability and age range); 2) intervention characteristics (intervention type and content, directed at individual or staff, and number of sessions); 3) number of participants, comparison groups, and design; and 4) outcome measures.
All included studies were assessed by the first author (EB) on potential sources of bias: random sequence, allocation concealment, blinding of outcome assessment, incomplete outcome data, selective reporting, group similarity at baseline, and personal or financial gain (Higgins & Green, 2008). Additionally, the next step was the comparison between the effect sizes of studies with a low risk of bias and the effect sizes of studies with a high risk of bias through subgroup analysis.

Data analyses
Because we assumed that the true effect would vary between studies, we used the random effect model to calculate the summary effect (Borenstein, Hedges, Higgins, & Rothstein, 2009) using the software Comprehensive Meta-Analysis (CMA) Version 2.0 (Borenstein et al., 2009). The summary effect was expressed as the overall standard difference in means (Cohen's d). A Cohen's d of 0.2 was considered small, 0.5 moderate, and 0.8 large (Cohen, 1988). We generated a forest plot of the overall random-effect of interventions and measured heterogeneity with I 2 . The percentage of I 2 describes the variability that is due to heterogeneity rather than sampling error (Borenstein et al., 2009). Values around 25% are considered low, 50% is considered moderate and 75% is considered high (Higgins, Thompson, Deeks, & Altman, 2003). To perform the random-effects metaanalysis, we held to the following assumptions: 1) if test-retest correlation of instruments was not specified in the paper, we used a correlation of r = 0.5; 2) if a study contained multiple parameters measuring different challenging behaviours, we used a summarised measure for the calculation of an overall challenging behaviours measure ('Use the mean of the selected outcomes' option of Comprehensive Meta-Analysis software); 3) if a studied intervention resulted in significant improvement of behaviours, the direction of the effect was stated positive; 4) data were standardised by post score standard deviations (SD); and 5) in case of multiple follow-up time points, these were computed together as a single measure. We conducted a sensitivity analysis using 'one study removed analyses' (Borenstein et al., 2009), to investigate the robustness of our results.
We performed four subgroup analyses to examine differences in treatment effects across assessment times, intervention types, delivery modes and study designs. For the first analysis, we compared post intervention assessments with follow-up assessments, to examine long-term effectiveness. Second, we categorised all included interventions into five intervention types, based on their theoretical background: 1) ABA or behavioural interventions, 2) cognitive behavioural therapy (CBT), 3) interventions combining mindfulness and behavioural techniques, 4) multisensory therapy, and 5) specialised teams using personalised treatment plans (i.e. Invididualized Habituation Plan (IHP)). All categories (i.e. intervention type) were compared on effectiveness. In the third and fourth subgroup analyses we compared interventions directly aimed at the individual with environment mediated interventions, and RCTs with non-RCTs, respectively.
Finally, to examine possible publication bias we generated a funnel plot (Duval & Tweedie, 2000) and used the Duval and Tweedie's trim and fill option to detect missing studies in the funnel plot.

Population
Dependent variable Actions to alter behaviour cognitive impair* behavio* AND problem* therap* mental* AND retard* tantrums treat* intellectual* AND disab* aggressi* interven* learning AND disab* self-inju* behavio* AND modification developmental* AND disab* self-inflicted AND wounds training adult Self-mutilation applied behavio* analysis elderly stereotyp* positive AND behavio* AND support individual challenging AND behavio* problem AND behavio* aggressive AND behavio* aberrant AND behavio* provocative AND behavio* stereotyped AND behavio* repetitive AND behavio* disruptive AND behavio* destructive AND behavio* maladaptive AND behavio* Journal of Intellectual Disability Research VOLUME 64 PART 8 AUGUST 2020

Study characteristics
The combination of the electronic search and reference tracking resulted in 10264 titles. After the three-step screening procedure 22 studies were included for this meta-analysis. The complete selection procedure is illustrated in Figure 1.
Together, the 22 included studies contained 1676 participants. An overview of all characteristics (i.e. participant characteristics, design, outcome parameters) and intervention characteristics (i.e. content, directed at individual or staff, number of sessions, intervention type) is presented in Table 2.
We solely used the keyworkers data of the Aberrant Behavior Checklist. Namely, the majority of included papers used keyworkers /staff members as informants.  Additionally, some data of the home carers were missing. Figure 2 shows an overview of the risk of different sources of bias of the included studies. Information on 'personal or financial gain' was too often missing to draw conclusions on. Additionally, due to limited variation in sources of bias between studies and frequent 'unclear' scores we had to refrain from comparing the effect sizes of studies with low risks of bias to the effect sizes of studies with high risk of bias.  Journal of Intellectual Disability Research VOLUME 64 PART 8 AUGUST 2020

Meta-analysis
The random-effects model showed an overall treatment effect with a moderate effect size (d = .573, P < .001, CI [0.352, 0.795]). The individual and combined effect sizes, lower limits, upper limits, z-values and p-values are presented in Figure 3. Heterogeneity was high (I 2 = 91.40%). The sensitivity analyses showed that the effect sizes varied between 0.491 and 0.666. These values fall within the range of the confidence interval of the overall effect size, indicating that our results were robust.

Subgroup analyses
We found no significant differences between post-intervention assessments versus follow-up assessments (Q = 0.198, d.f. = 1, P = 0.656). There was however a significantly higher effect of interventions combining mindfulness and behavioural techniques than of all other intervention types (Q = 9.176, d.f. = 1, P = 0.002). There were no

Publication bias
The funnel plot (Figure 4) shows clear asymmetry, with a predominance of papers on the right range of the plot (displayed as white dots in Figure 4), suggesting publication bias. The unequal distribution of effect sizes of our included studies was confirmed by the Duval and Tweedie's trim and fill analysis. The eight black dots on the left side of the plot represent expected studies with negative effect sizes that were not included in the meta-analysis. This finding suggests that there may have been studies that have not been published.

Discussion
This meta-analysis provides insight in the effectiveness of non-pharmacological interventions to treat challenging behaviours in adults with intellectual disabilities. We found a moderate overall effect of non-pharmacological interventions, consistent with some previous meta-analyses (Harvey et al., 2009;Heyvaert et al., 2010;Shogren et al., 2004). However, some other reviews and meta-analyses did not find evidence for the effectiveness of non-pharmacological interventions (Gustafsson et al. 2009;Hassiotis & Hall, 2008;Chan et al., 2010;. This difference in findings may be due to the different aims of previous reviews and meta-analyses. For example, the meta-analysis of  was specifically aimed at cognitive behavioural treatment for anger in adults with challenging behaviours and intellectual disabilities, while the meta-analysis of Heyvaert et al. (2010) more broadly examined pharmacological, psychotherapeutic, and contextual interventions for treating challenging behaviours in individuals with intellectual disabilities.
Our results indicate that effect sizes of non-pharmacological interventions are also moderate effective on the long-term (follow-up measures ranged from 3 to 18 months), suggesting that treatment effects of non-pharmacological interventions sustain after the intervention has ended. However, we must be cautious with the interpretation and implications of this finding, as the measures of post intervention assessments and follow-up assessments are not independent from eachother. To our knowledge, there have been no earlier studies that have compared post-intervention effects with follow-up effects. Currently, in clinical practice, pharmacological treatments, instead of non-pharmacolocial interventions, are often the first treatment of choice (Holden & Gitlesen, 2004). This may be due to the immediate effects of medication, in contrast to the gradual effects of non-pharmacological interventions (Beadle-Brown, Mansell, Whelton, Hutchinson, & Skidmore, 2006). Moreover, non-pharmacological interventions often require a substantial time investment of health care professionals (Matson & Wilkins, 2008). However, the use of medication is controversial due to negative side effects (Matson & Mahan, 2010;Sheehan et al., 2017) and questionable effectiveness (Scheifes et al., 2016;Shankar, Wilcock, Oak, McGowan, & Sheehan, 2019;Sheehan et al., 2015). The possible long-term positive outcomes we found of non-pharmacological interventions might motivate clinicians to invest in non-pharmacological interventions more often, rather than medication.
Interventions combining mindfulness with behavioural techniques showed to be more effective than behavioural interventions without mindfulness components, CBT, multisensory therapy, and individualised habituation plans. No previous studies have demonstrated the superiority of this type of interventions (Heyvaert et al., 2010;Hwang & Kearney, 2013). However, this finding should be interpreted with care. Subgroup analyses may be misleading, due to missing randomised comparisons, which makes the results more susceptible to false positive tests results (Higgins & Green 2008). Moreover, all included studies reporting on interventions combining mindfulness and behavioural techniques came from the same research group. RCTs from other research groups, with head to head comparisons, are necessary to draw more robust conclusions on the effects of those interventions on challenging behaviours of adults with intellectual disabilities.
We found no differences in effect between individual directed interventions and environment mediated interventions. Earlier reviews and meta-analyses demonstrated that interventions that were aimed at altering the environment, or at training staff were effective (Brosnan & Healy, 2011;Heyvaert et al., 2010;Heyvaert et al., 2012), while other reviews and meta-analyses did not (Cox et al., 2015;Knotter et al., 2018;van Oorsouw, Embregts, Bosman, & Jahoda, 2009). Our results indicate that there are no differences in effect sizes between interventions aimed at the environment versus at the individual. However, there are clear differences in applicability of individual directed interventions versus environment mediated interventions. For instance, to conduct CBT, the individual needs the verbal skills to express feelings and thoughts (Sturmey, 2004) which is only the case in higher functioning individuals with intellectual disabilities. In contrast, environment mediated interventions, such as staff training, are more broadly applicable to individuals with different levels of intellectual disabilities. Such interventions provide staff with tools that they can use more consistently, and apply in new situations, possibly indicating a more sustainable effect. However, implementing such environment mediated interventions is known to be a struggle (Bosco et al., 2019). Insufficient training and supervision, high turnover rates, time constraints and low support from management have shown to be pitfalls in implementing environment mediated interventions (Bosco et al., 2019;Campbell, 2010). As a consequence, the risk of ineffective treatment increases (Feldman, Atkinson, Foti-Gervais, & Condillac, 2004).
In line with the meta-analysis of Heyvaert et al. (2010), we did not find differences in effect sizes between RCTs and non-RCTs, indicating no evidence for overestimation of treatment effect of non-RCTs. It is interesting however, that the number of RCTs in the field on non-pharmacological intervention studies appears to be rising. Previous reviews and meta-analyses reported a scarcity of methodologically sound clinical trials in the field on non-pharmacological intervention studies for adults with intellectual disabilities and challenging behaviours (Gustafsson et al. 2009;Hassiotis & Hall, 2008;. In our meta-analysis, the balance between RCTs (n = 11) and non-RCTs (n = 11) was more even than in earlier ones (Heyvaert et al., 2010; including 5 RCTs against 10 non-RCTs;  including 2 RCTs against 10 non-RCTs). The increasing number of RCTs is promising, especially because conducting clinical trials in the field of non-pharmacological intervention studies for adults with intellectual disabilities and challenging behaviours is known to be challenging (Cleaver et al. 2010;Robotham et al. 2011;Nicholson, Colyer, & Cooper, 2013). Many clincial trials experienced recruitment problems, high drop out rates and high staff turnover (Bhaumik, Gangadharan, Hiremath, & Russel, 2011;Hassiotis et al., 2018). Only recently was the first paper on process evaluation of a non-pharmacological intervention study (e.g. positive behaviour support) published (Bosco et al., 2019), showing that participants found it difficult to combine trial required assessments with routine clinical care. More of these process evaluations are warranted, as they increase insight in the specific barriers of conducting clinical trials in the field of adults with intellectual disabilities. Findings may help prevent such problems for future studies or to apply more flexible trial designs.
Previous studies indicated that interventions applying functional analysis were more effective than interventions which did not incorporate this (Didden et al., 2006;Harvey et al., 2009;Brosnan & Healy, 2011;Heyvaert et al., 2012;Lydon et al. 2013;Lloyd & Kennedy 2014). Moreover, the use of functional analysis is recommended by clinical guidelines (Banks & Bush, 2016; National Institute for Health and Care Excellence, 2019). Unfortunately, in our meta-analysis we were unable to analyse whether intervention effects differed in this respect, as some of the included papers were ambiguous about the incorporation of assessment of function in the intervention. Important to note is the high heterogeneity we found, which indicated that most of the observed variance was real. However, our sensitivity analysis showed that the effect sizes all stayed within the range of the confidence interval of the overall moderate effect size, indicating that our results were robust. Moreover, we anticipated that the true effect sizes would vary. Hence we conducted a random effect model, which is more conservative than the fixed effect model (Fletcher, 2007). Furthermore, the overall effect of our study is in line with previous, broad aimed meta-analyses which compared wide ranges of interventions (Didden et al., 2006;Heyvaert et al., 2010;Heyvaert et al., 2012). Therefore, we believe that our results are a valuable addition to the body of evidence on the effectiveness of non-pharmacological interventions. Another important finding of our study was publication bias we found. Our results showed that especially large scale trials reporting no or negative effects were missing. Some previous meta-analyses also detected publication bias (Denis, van den Noortgate, & Maes, 2011;Heyvaert et al., 2012), while others did not (Hart & Banda, 2010;Heyvaert et al., 2010;Knotter et al., 2018). Since we only included English-written papers, we expected a certain level of publication bias. In the future, consequent registration of trials is important to bring about more transparency on studies and reduce publication bias.
The strength of our study was the number of studies that conducted large scale RCTs. However, our findings should also be interpreted in light of its limitations. A first limitation was the exclusion of eligible studies due to missing data or missing papers. Despite our efforts to collect all necessary data and papers, we could not get in touch with some authors, or the authors could not provide us with the necessary data, and we therefore had to exclude their studies (n = 12). The exclusion of approximately a third of the eligible papers increased the risk of bias and may have affected our results. Second, we did not include single-case studies in our meta-analysis. This resulted in a loss of papers, especially from earlier research on interventions for challenging behaviours within the population of individuals with intellectual disabilities. However, we chose to include only studies with experimental or quasi-experimental designs, in order to update and build upon previous meta-analyses of studies using these kind of designs (Heyvaert et al., 2010;Knotter et al., 2018). This approach also had the advantage of being able to analyse a methodologically more homogeneous group of studies, compared to meta-analyses including small-n designs as well (e.g., . Third, 'non-pharmacological interventions' could have been a too broad range of different interventions to cluster together for an overall effect, and indeed our results showed high heterogeneity. However, as previous stated, past meta-analyses have also included a broad range of interventions (e.g. Didden et al., 2006;Heyvaert et al., 2010;Heyvaert et al., 2012), which enhances the comparablility of our study with these studies. Fourth, the population that we examined (i.e. adults with intellectual disabilities) was quite heterogeneous. We included studies on individuals with all levels of intellectual disability (profound to borderline) and a broad age range. Unfortunately, we were unable to collect individual participant data (i.e. level of intellectual disability and age) of the included studies, therefore we could not analyse the effect of these characteristics on intervention effects and heterogeniety. Future effectiveness studies should focus on how and which participant characteristics affect treatment success (i.e. level of intellectual disability, age). Finally, we only examined the reduction of challenging behaviours as a measure of treatment success. While challenging behaviours have far reaching negative consequences, for the individuals with intellectual disabilities as well as their environment, future studies should take quality of life of the individual with intellectual disabilities, or emotional wellbeing of staff into account as other relevant parameters in the evaluation of treatment effectiveness.
In conclusion, we found a moderate effect of non-pharmacological interventions in reducing challenging behaviours in adults with intellectual disabilities, and this effect appears to be longlasting. To assess the superiority of different types of interventions, more research is needed. Fortunately, there is a positive development in the scientific field with the growing numbers of large scale, RCTs that are being conducted. For future research, trial registration and conducting more large scale studies with high quality designs is necessary. Furthermore, future studies should examine the effect of participant characteristics on treatment success, such as level of intellectual disability and age, and take other outcome measures into account, such as quality of life or staff wellbeing. These steps will add to a more comprehensive perspective on the effect of non-pharmacological interventions.