Elsevier

Research in Autism Spectrum Disorders

Volume 5, Issue 1, January–March 2011, Pages 592-603
Research in Autism Spectrum Disorders

Predictors of outcome for children receiving intensive behavioral intervention in a large, community-based program

https://doi.org/10.1016/j.rasd.2010.07.003Get rights and content

Abstract

This study reports on predictors of outcome in 332 children, aged 2–7 years, enrolled in the community-based Intensive Behavioral Intervention (IBI) program in Ontario, Canada. Data documenting children's progress were reported in an earlier publication (Perry et al., 2008). The present paper explores the degree to which four predictors (measured at intake to IBI) are related to children's outcomes: age at entry, IQ, adaptive scores, and autism severity. Outcome variables examined include: post-treatment scores for: autism severity, adaptive behavior, cognitive level, rate of development in IBI, and categorical progress/outcomes (seven subgroups). All four types of predictors were related to children's outcomes, although initial cognitive level was the strongest predictor. In addition, two subgroups of the sample are examined further. Children who were most successful in the program and achieved average functioning had higher developmental levels at intake, were considerably younger than the rest of the children, and were in treatment longer than children in other outcome categories. Children who were least successful in the program and made essentially no progress did not differ appreciably from the remainder of the group. Implications of these results for decision-making are discussed.

Introduction

Early Intensive Behavioral Intervention (EIBI or, as typically designated in Ontario, IBI) is an intensive application of the principles of Applied Behavior Analysis designed for young children with autism. The intervention is comprehensive in scope and is typically provided in a 1-to-1 format (at least initially) for 20–40 h per week for about 2 years. It is intended to change children's developmental trajectory such that they can transition to learning in a more typical way in school (Lovaas, 1987, McEachin et al., 1993). The behavioral nature of the intervention, not just its intensity, appears to be linked to superior outcomes (Eikeseth et al., 2002, Howard et al., 2005). Although IBI is now widely regarded as best practice for children with autism, outcomes are decidedly variable with good outcomes for about half the children at best. Although this is remarkable compared to the results of autism intervention prior to the widespread use of IBI, clearly IBI is not a panacea for all children. Research, to date, has not been able to account very well for this heterogeneity in outcomes.

Several investigators have focused on examining more closely which children make very dramatic gains and achieve “recovery”, “best outcomes” or average functioning (e.g., Eikeseth et al., 2007, Lovaas, 1987). However, it is also important to investigate further which children do not benefit, which has rarely been done. As noted by Reichow and Wolery (2009), “…it is imperative children not responding to intervention are identified early so additional and/or different treatments can begin.” (p. 39). Considerations regarding resource allocation in the face of limited budgets and active advocacy efforts loom large. It is clear that greater understanding is urgently needed regarding which children will benefit, to what degree, and why.

Questions regarding moderators or predictors of outcome have been increasingly addressed in the literature recently. Factors that have been theorized to be related to the heterogeneity in outcome include child, family, and treatment characteristics. The majority of research, including the present study, focuses on child characteristics, such as cognitive and adaptive levels, age at IBI onset, and severity of autism symptomatology. However, it should be noted that some consideration is also being given recently to treatment characteristics (Granpeesheh et al., 2009, Koudys and Perry, 2010, Makrygianni and Reed, 2010, Reichow and Wolery, 2009) and family variables (Remington et al., 2007, Shine and Perry, 2010, Solish, 2010).

There are good reasons to assume that starting IBI younger might be beneficial and this could be argued from several different theoretical perspectives (e.g., behavioral theory, neural plasticity, and developmental theory). A number of studies have reported on the question of children's age when treatment commences and whether the widespread “earlier the better” belief is empirically borne out. Results are somewhat equivocal, in fact. Studies which have wide age ranges and have divided their sample into younger versus older subgroups have typically found that younger (however defined; under 4 or under 5) children are more likely to show better outcomes than older children (Fenske et al., 1985, Granpeesheh et al., 2009, Harris and Handleman, 2000) and the same is true of Anderson, Avery, DiPietro, Edwards, and Christian (1987) based on the individual data in the paper. However, other studies (typically using correlational type statistics) have looked for and not found a relationship with age. This is true in very young samples (2–3.5 years; Hayward et al., 2009, Lovaas, 1987) and also one somewhat older sample (4–7 years; Eikeseth et al., 2002, Eikeseth et al., 2007). Small samples with restricted age ranges may preclude correlations from emerging as significant in these studies. Other studies appear not to have examined the question of age at all, or at least do not report such analyses. Thus, as noted by Matson and Smith (2008), the precise nature and power of age as a predictor remains less clear than one might think.

Children's initial cognitive level has also been examined as a predictor of outcome in a number of studies. Initial IQ has often been reported to be moderately to highly correlate with outcomes (Eikeseth et al., 2002, Eikeseth et al., 2007, Harris and Handleman, 2000, Hayward et al., 2009, Sallows and Graupner, 2005). However, this is likely the case regardless of treatment as exemplified by results reported by Gabriels, Hill, Pierce, Rogers, and Wehner (2001) for more generic treatment and the Eikeseth et al. (2007) eclectic comparison group. Although most studies find initial IQ related to outcome, a few studies have examined this relationship and found it not to be significant (Birnbrauer and Leach, 1993, Cohen et al., 2006, Smith et al., 2000).

Although adaptive behavior measures are frequently used as outcome measures, less attention has been paid to initial adaptive levels as predictors of outcome but there is some evidence that children with better adaptive skills tend to have better outcomes (Remington et al., 2007, Sallows and Graupner, 2005). Since cognitive and adaptive levels are correlated, one might expect similar results, but because of the interesting relationship between cognitive and adaptive scores at different cognitive levels (Perry, Flanagan, Dunn Geier, & Freeman, 2009), it might be worthwhile examining adaptive levels as a predictor of outcomes in a heterogeneous sample.

Surprisingly, severity of autism symptoms or diagnosis (Autism versus PDD-NOS) has rarely been included in predictor analyses or even as an outcome measure (Matson, 2007, Matson and Smith, 2008). Sallows and Graupner (2005) showed that lower pre-treatment ADI-R scores (together with higher IQ and more rapid early skill mastery) were quite accurate in predicting which children would be “rapid responders”. Remington et al. (2007), on the other hand, found that their good responders initially had more severe autism symptoms. Smith et al. (2000) found that their Autism subgroup showed only a 4-point IQ gain versus a 16-point IQ gain in the PDD-NOS subgroup.

In summary, for all four child factors (initial age, cognitive level, adaptive skills, and diagnostic severity), results are not completely consistent and further research is needed. However, most of these studies are quite small (e.g., 10–25 children) and, thus, power limitations preclude many analyses which might shed light on why some children do better than others. The small samples likely result in Type II error (i.e., interesting findings may have been missed). Also, possible sampling issues could result in spurious findings as well, as some of the samples are quite restricted in range (e.g., the children in Lovaas [1987] were all less than 3.5; the children in Eikeseth et al. (2002) all had IQs over 50). Therefore, systematic and meta-analytic studies of this body of literature have started to appear, which seek to provide greater clarity by systematically combining the existing literature.

Howlin, Magiati, and Charman (2009) conducted a systematic review, identifying 11 studies which met their criteria. They concluded that IBI resulted in improved outcomes at a group level but that there was considerable individual variability. In terms of predictors, they concluded, based on their descriptive summary of the studies, that initial IQ and receptive language were important predictors of IQ at follow-up but that initial age and diagnosis were unrelated to outcomes. Eldevik et al. (2009) used a more specific criterion for inclusion of studies which resulted in examining nine controlled studies. They calculated effect sizes for IQ and adaptive behavior and conducted a meta-analysis of individual children's data from all studies combined. Results indicated a large effect size for IQ (1.10) and medium effect size for adaptive behavior (.66). However, Eldevik et al. (2009) did not conduct predictor analyses.

Reichow and Wolery (2009) conducted a systematic review of 13 studies, comprising a total of 251 children receiving IBI, which included ratings of methodological rigor, participant characteristics, intervention quantity and quality. They computed both mean change effect sizes (comparing pre- and post-treatment scores) and mean difference effect sizes (comparing the treatment group with a control or comparison group) for IQ, adaptive behavior, expressive language, and receptive language, and conducted a meta-analysis, using corrections for small sample sizes and other procedures to ensure conservative conclusions. Of greatest relevance to the present paper, Reichow and Wolery (2009) also conducted moderator analyses, examining the studies’ weighted effect sizes for IQ (the main dependent variable) as a function of several child and treatment variables. Model of supervision was the only significant moderator across studies with the UCLA-model supervision associated with superior effect sizes. Other treatment variables reflecting amount of treatment (intensity and duration) were unrelated to child outcome, at least within the ranges studied. Neither of the child characteristics analyzed, that is, pre-treatment age and pre-treatment IQ, was found to significantly affect children's outcomes (post-treatment IQ). Diagnostic severity was not examined because most studies did not include it.

Most recently, Makrygianni and Reed (2010) conducted a meta-analysis of 14 studies, again examining both mean change effect sizes and mean difference effect sizes, but using slightly different procedures, for IQ, language, and adaptive behavior. They also examined the relationship of effect sizes with program variables (intensity, duration, and parent training) and child characteristics (age at intake, initial levels of cognitive, language, and adaptive skills), controlling for methodological quality of the studies. Their results were not necessarily consistent with those of Reichow and Wolery (2009). For example, they found intensity of treatment to be significantly related to several of the effect sizes. Their analyses for age at intake revealed moderately high negative correlations for age at intake with several of their effect size variables. Interestingly, a scatter plot of effect sizes for all variables by age at intake suggested that studies with children who began very early (roughly under 3) tended to have more uniformly large effect sizes whereas studies with children beginning later had more variable effect sizes. In terms of developmental level as a predictor, Makrygianni and Reed (2010) found that intellectual ability at intake was not correlated with effect sizes but that intake IQ was very highly correlated with outcome IQ. Initial adaptive behavior was related to effect sizes for language outcomes and adaptive behavior outcomes. Autism severity was not examined in this study (again because of limited data).

The purpose of the present paper was to report on our analyses of predictors in a very large (n = 332) and diverse sample of children receiving IBI. The sample is larger than the combined sample used by the authors of the meta-analyses and it is very diverse in terms of children's developmental and diagnostic characteristics as well as family background. As such, there is sufficient range and statistical power to explore prediction of outcomes more fully than has been possible before. Thus, we hoped that these analyses would help to shed some light on the debate in the literature regarding the relative importance of different child predictor variables. In particular, we set out to examine the relationship of children's outcomes to age at entry, initial cognitive level, adaptive functioning, and diagnostic severity. In addition, we examined correlates and predictors more closely in two subgroups of children at the two extremes of outcome classification: those with outcomes in the average range and those with poor outcomes.

Data for this paper are drawn from the effectiveness study of the Ontario province-wide IBI initiative, a large, community-based publicly funded IBI program. The earlier paper (Perry et al., 2008) addressed the question of whether children showed statistically significant and clinically significant improvement on developmental and diagnostic measures and describes the range of progress/outcomes seen in children. Briefly, results indicated that, overall, children improved significantly across all measures from entry (T1) to exit (T2) but there was substantial heterogeneity. More specifically, they demonstrated significantly milder autistic symptomatology at T2. Adaptive behavior age equivalents increased substantially in all areas, but standard scores changed only modestly (higher for Socialization and Communication but lower for Daily Living Skills). Cognitive level (when available) increased significantly. Children's rate of development during IBI was roughly double what it had been at T1. Based on a combination of all available measures, children were classified as falling into one of seven categories of progress/outcome: Average, Substantially Improved, Clinically Significantly Improved, Less Autistic, Minimally Improved, No change, and Worse. A subgroup of children who were more similar to children from model programs (younger with milder developmental delays) had similar outcomes to those reported in efficacy studies.

Section snippets

Participants

Psychological assessment file data were available for a total of 332 children (83% boys) with an entry assessment (T1) and another assessment (T2, usually at exit). More details about the sample may be found in the earlier report (Perry et al., 2008). The children's initial status on all diagnostic and developmental variables is shown in Table 1, as well as their age (which was, on average, about 4.5) and the duration of intervention (which was, on average, 18 months).

Measures

Assessment measures for

Age at entry

Do children beginning IBI earlier show greater gains? This question was examined in several different ways. First, as shown in Table 2, the adaptive and cognitive variables at outcome tended to be significantly negatively correlated with age at entry (i.e., children who started IBI younger tended to score higher at discharge), and younger age at entry was correlated with milder autism severity at exit. These correlations are small to medium in magnitude.

Second, data were compared using

Discussion

This paper reports on initial age at entry, cognitive level, adaptive skills, and diagnostic severity as predictors of outcome in a large group of children receiving community-based IBI (the same group whose outcomes were reported in Perry et al., 2008). All four predictors were found to be important, to some extent, in relationship to outcomes of cognitive and adaptive level, severity of autism, rate of development during intervention, and categorical outcome.

Earlier age at program entry was

Acknowledgements

We are grateful to the Ontario Ministry of Children and Youth Services for funding this study. However, the views expressed here are those of the authors and do not necessarily represent the position of the Ministry. The Ministry in no way influenced the interpretation or writing of this paper. We appreciate the assistance of Alissa Levy, Helen Penn Flanagan, Alice Prichard, Abbie Solish, April Sullivan, and Kerry Wells with data collection, data entry and verification and Don Downer for data

References (36)

  • H. Cohen et al.

    Early intensive behavioral treatment: Replication of the UCLA model in a community setting

    Developmental and Behavioral Pediatrics

    (2006)
  • S. Eikeseth et al.

    Intensive behavioral treatment at school for 4–7-year-old children with autism: A 1-year comparison controlled study

    Behavior Modification

    (2002)
  • S. Eikeseth et al.

    Outcome for children with autism who began intensive behavioral treatment between ages 4 and 7: A comparison controlled study

    Behavior Modification

    (2007)
  • S. Eldevik et al.

    Meta-analysis of early behavioral intervention for children with autism

    Journal of Clinical Child and Adolescent Psychology

    (2009)
  • Freeman, N. L., Brown, R., Dunn Geier, J., Lindblad, T., Perry, A., & Reitzel, J., et al. (2008). The development of...
  • Flanagan, H. E. (2009). The impact of community-based Intensive Behavioural Intervention. Unpublished doctoral...
  • R.L. Gabriels et al.

    Predictors of treatment outcome in young children with autism

    Autism

    (2001)
  • R.P. Goin-Kochel et al.

    Early responsiveness to intensive behavioral intervention predicts outcomes among preschool children with autism

    International Journal of Disability, Development and Education

    (2007)
  • Cited by (94)

    • Early autism analysis and diagnosis system using task-based fMRI in a response to speech task

      2021, Neural Engineering Techniques for Autism Spectrum Disorder: Volume 1: Imaging and Signal Analysis
    View all citing articles on Scopus

    Portions of this study were presented in poster or presentation formats at: the Association on Behavior Analysis Autism Conference in February 2009 in Jacksonville, FL; the Association for Behavior Analysis in May 2007 in San Diego, CA; the Society for Research in Child Development in April 2007 in Boston, MA; the Ontario Association on Developmental Disabilities in April 2007 in Barrie, ON; the Association on Behavior Analysis Autism Conference in February 2007 in Boston, MA; and at the Ontario Association for Behaviour Analysis in November 2006 in Markham, ON.

    View full text