**5. Evidence base**

The evidence for ABA-based interventions spans all valid and recognised research method‐ ologies, including Single-System Design (SSD), Randomised Controlled Trials (RCT), Metaanalysis and Sequential Meta-analysis, Systematic Reviews, Social Validity studies, Neuroscience studies, and Cost-benefit analysis.

*Single System Designs* (SSD) include reversal designs, multiple baseline designs (across behaviours, settings, or subjects), changing criterion designs, and alternating treatment designs [46]. In SSD studies internal validity is achieved by each participant serving as his/her own control, while external validity/generality is achieved through numerous replications of carefully described SSD methodologies.

Hundreds, if not thousands, of Single-System Design (SSD) studies have been published evidencing the effectiveness of ABA for individuals with autism [22]. While most of these studies are published in flagship journals, such as *Journal of Applied Behavior Analysis,* increas‐ ingly other mainstream journals publish SSD evidence for ABA-based interventions, for example, the *British Journal of Special Education* [15].

A good example of an SSD is Garcia-Albea, Reeve, Brothers, and Reeve (2014), who used a multiple-probe design across participants to teach 4 boys with autism to initiate and participate in social interactions without vocal prompts from adults. The procedure involved the use of a script and script-fading procedure. The boys quickly learned to talk independently about a whole range of relevant things in their environment without the help of adults. While this kind of research methodology lends itself particularly well to the action-based researcher/scientist-practitioner model inherent in ABA, it can be usefully employed in a range of different settings [49].

*Randomised controlled trial* (RCT), sometimes held up as the 'gold standard' for evidence of effectiveness of interventions, originated from medical research. RCTs were developed to compare outcomes for one group of people who receive a certain type of medication (treatment group), to that of another group of people who are not receiving the same medication, i. e., who may have received a placebo or 'treatment as usual' (control group). The basic assumption underpinning RCT is that, if both groups of people are well matched, any differences that are observed after the intervention are due to the intervention [37].

While RCTs may have their utility in relatively clear-cut medical research, there are many problems when they are used in social care or educational research, not least the ethical dimension of withholding a potentially beneficial treatment from the control group. Of course, there are safeguards, such as cross-over designs or the Hippocratic Oath to 'do no harm' [90].

However, some of the main drawbacks in autism research are that, for RCT results to be valid, all members of the 'treatment group' have to receive the exact same treatment and this has to be held stable for the agreed duration of the intervention. Of course, when interventions are based on a functional analysis of behaviour, as is the case in ABA, they are tailored to the needs of the individual, i. e., they are person/child-centered. Data-based decisions are made with regards to intervention adjustments, that are implemented immediately, for ethical reasons, in order to avoid harm and enhance treatment effects [14]. These kinds of progressive, systematic, individualized, data-based intervention revisions and adjustments would inva‐ lidate RCT data (see Single-System Designs).

**5. Evidence base**

250 Autism Spectrum Disorder - Recent Advances

Neuroscience studies, and Cost-benefit analysis.

example, the *British Journal of Special Education* [15].

employed in a range of different settings [49].

observed after the intervention are due to the intervention [37].

carefully described SSD methodologies.

The evidence for ABA-based interventions spans all valid and recognised research method‐ ologies, including Single-System Design (SSD), Randomised Controlled Trials (RCT), Metaanalysis and Sequential Meta-analysis, Systematic Reviews, Social Validity studies,

*Single System Designs* (SSD) include reversal designs, multiple baseline designs (across behaviours, settings, or subjects), changing criterion designs, and alternating treatment designs [46]. In SSD studies internal validity is achieved by each participant serving as his/her own control, while external validity/generality is achieved through numerous replications of

Hundreds, if not thousands, of Single-System Design (SSD) studies have been published evidencing the effectiveness of ABA for individuals with autism [22]. While most of these studies are published in flagship journals, such as *Journal of Applied Behavior Analysis,* increas‐ ingly other mainstream journals publish SSD evidence for ABA-based interventions, for

A good example of an SSD is Garcia-Albea, Reeve, Brothers, and Reeve (2014), who used a multiple-probe design across participants to teach 4 boys with autism to initiate and participate in social interactions without vocal prompts from adults. The procedure involved the use of a script and script-fading procedure. The boys quickly learned to talk independently about a whole range of relevant things in their environment without the help of adults. While this kind of research methodology lends itself particularly well to the action-based researcher/scientist-practitioner model inherent in ABA, it can be usefully

*Randomised controlled trial* (RCT), sometimes held up as the 'gold standard' for evidence of effectiveness of interventions, originated from medical research. RCTs were developed to compare outcomes for one group of people who receive a certain type of medication (treatment group), to that of another group of people who are not receiving the same medication, i. e., who may have received a placebo or 'treatment as usual' (control group). The basic assumption underpinning RCT is that, if both groups of people are well matched, any differences that are

While RCTs may have their utility in relatively clear-cut medical research, there are many problems when they are used in social care or educational research, not least the ethical dimension of withholding a potentially beneficial treatment from the control group. Of course, there are safeguards, such as cross-over designs or the Hippocratic Oath to 'do no harm' [90].

However, some of the main drawbacks in autism research are that, for RCT results to be valid, all members of the 'treatment group' have to receive the exact same treatment and this has to be held stable for the agreed duration of the intervention. Of course, when interventions are based on a functional analysis of behaviour, as is the case in ABA, they are tailored to the needs of the individual, i. e., they are person/child-centered. Data-based decisions are made with Of course ABA (i. e., the application of the scientific discipline of behaviour analysis) itself cannot be assessed via RCTs, yet some specific intervention packages, such as Early Intensive Behaviour Interventions (EIBI) or the Early Start Denver Model, have been assessed in RCTs. A good example is Howard, Sparkman, Cohen, Green, and Stanislaw [39], who evaluated 29 pre-school children who received intensive behaviour analytic intervention (treatment group) and two matched control groups of 16 children each, receiving either intensive or non-intensive ''eclectic'' interventions. While the scores for cognitive, language, and adaptive skills were similar at intake, at follow-up the treatment group had statistically significant higher mean standard scores in all areas. These data were confirmed at the 2 year follow-up [40].

Other RCTs or quasi-experimental control studies have compared Treatment as Usual with ABA-based interventions, such as specific commercially available intervention packages [34), high vs low intensity ABA-based interventions [30, 59], or waitlist controls [67].

*Meta-analysis* and *sequential meta-analysis* are increasingly used to give a summary of multiple small n studies that provide individual participant data, with the expectation that combining these data (commonly calculated in effect sizes) will allow for the identification of patterns and thus increase statistical powerto show that treatment effects are not due to measurement error, variation in sample, etc. *Sequential meta-analyses* are conducted where enough cumulative knowledge is available through meta-analysis to draw convincing statistical conclusions about effect size. Of course as in all research, there are a number of issues related to researcher bias and declaration of interest, however, over recent years meta-analyses have become a welcomed addition to the evidence-based practice literature.

With regards to autism interventions, a recent overview of meta-analyses [77] found that early intensive ABA-based treatment was significantly related to enhanced outcomes (effect sizes 0. 30 to > 1). Further meta analyses [22, 23, 24, 72] and a recent sequential meta-analysis [54] have confirmed these findings [1].

*Systematic reviews* are based on detailed searches of data banks with clearly defined inclusion/ exclusion criteria. Usually teams of multidisciplinary experts summarise selected studies, such as RCTs, single-system research design studies, and meta-analyses. Given the wide reach of evidence covered in systematic reviews, they have gained a strong place in evidence-based practice in ASD.

The number of systematic reviews of ASD interventions has risen recently [77]. By-and-large ABA-based interventions, in particular Early Intensive Behavioural Interventions (EIBI), are endorsed by systematic reviews. A good example of a comprehensive systematic review was carried out by the large scale multidisciplinary team of the National Autism Center [63] ; 11 interventions were designated as established, of these all but one are explicitly based on ABA; 22 intervetions were categorised as emerging, most of these were also based on ABA. All other systematic reviews came to similar conclusions [6, 36, 70, 75, 91].

The review by [43] is the notable exception, in that it does not fully concur with these conclu‐ sions. Howlin et al. concluded that 'this review provides evidence for the effectiveness of EIBI for some, but not all, preschool children with autism' (p. 20). Given that this review is fre‐ quently cited in the UK as a basis against the roll-out of EIBI for all children with ASD who need it [42], it is important to note here that Howlin et al. misinterpret a number of important points. First, it is in the mathematical nature of all group average data (such as those calculated for RCTs) that some individual data are above while others are below the average; such is the nature of group averages (see also [77]; second, Howlin et al. 'cherry pick' results by ignoring the fact that obviously some children must do extremely well, otherwise the group average would not be what it is. Thus, Howlin et al. contradict themselves in their conclusions. First they call for large sample comparisons and group averages (i. e., RCTs) and then they do not accurately interpret group data.

In a subsequent paper, Howlin and colleagues [99] report extremely poor long-term outcomes in a 40-year follow-up study of children diagnosed with autism at the Institute of Psychiatry/ Maudsley Hospital, London between 1950 and 1979. Intriguingly, they explicitly link these findings to the fact that none of these children had received early intensive behavioural interventions and claim that EIBI is available now. Praising the potential positive effects of EIBI stands in contrast to their earlier conclusions [42, 43]. It will be interesting to see how this new evidence will translate into advice given to government bodies.

Given that group average scores are neither sensitive to individual differences nor offer sufficient generality, most behaviour analytic researchers prefer to rely on replicated singlesystem designs (SSD) instead of group averages [14, 18, 29]. Clearly, SSD research data cannot be ignored and should find their rightful place in future reviews of autism intervention guidelines, such as NICE Guideline 170 [64].

*Social Validity* studies assess the social significance, appropriateness, and importance of treatment goals, procedures, and intervention effects [93]. Social validity measures are increasingly becoming integral part of research into interventions in ASD [27, 53].

A number of studies have shown clear evidence of high social validity of ABA-based inter‐ ventions, especially those that include parent participation and training [18, 92]. Interestingly, while there is evidence of increased parental stress in families affected by ASD [10, 17], there is evidence of parental stress reduction when effective interventions for children are in place [17]. This is also true for education staff [26].

*Neuroscience* studies, including MRI scans are useful tools to bolster evidence-based practice in particular in the area of ASD, where the plasticity of the brain during early childhood constitutes an important focus of intervention [11]. There is evidence of differences in brain activity between individuals diagnosed with ASD and those who do not have an ASD designation [13, 35].

There is further evidence that early behaviour analytic intervention can lead to measurable change in brain activity [12]. For example, [28] found that ABA-based interventions not only lead to behavioural improvements, with some optimal outcome individuals becoming 'indistinguishable' from neuro-typical peers, but that they also lead to improved neurological development, i. e., neurological plasticity allowing for compensatory development.

*Cost-benefit analyses* are an important way to substantiate evidence of effective interventions. A recent study estimated the annual 'cost of autism' between £0. 8-1. 4+ million per lifetime depending on the level of functioning; these costs were similar in UK and USA [7] and in other parts of the world [61, 71].

There is evidence that effective ABA-based interventions can reduce this cost substantially in the long-term, i. e., \$1+million per year [45]. However, due to the fact that intensive interven‐ tions generally are rather costly in the short-term, there has been resistance to their imple‐ mentation. The key question is how effective high-quality programs can be delivered in a more cost-effective sustainable model, without losing out on effectiveness [1].

All of these studies supply ample evidence of the effectiveness and efficacy of ABA-based interventions, in achieving individual potential in a full range of areas, including intellectual, social, and verbal, functioning, ASD symptomatology, and challenging behaviour.

On the basis of this evidence, ABA-based interventions are now widely endorsed in the USA, Canada, Australia, and some European countries. On a federal level in the USA, for example, Medicaid now covers ABA-based interventions and the Affordable Care Act covers behav‐ ioural health treatments [83], which include ABA-based interventions generally, and is not restricted to ASD diagnosis.
