Randomized, Controlled Trial of an Intervention for Toddlers With Autism: The Early Start Denver Model

Dear Editor:

Dawson and colleagues published a highly important study which provides much-needed evidence for the efficacy of early behavioral interventions for preschoolers with autism spectrum disorders under the age of 30 months. There has been much speculation and belief that the earlier the intervention, the better the outcome, but robust data to support this assumption were thus far lacking. This study therefore fulfills a knowledge gap that will have wide implications for service providers worldwide. Not only will the study provide an evidence-based treatment for preschoolers who are diagnosed at increasingly earlier stages, but it will also inform practitioners on interventions that are required for children who present atypical developmental trajectories without meeting full criteria for ASD, such as siblings of autistic probands who are at risk and are now frequently followed closely from an early age.

The strengths of this study are numerous. The authors have combined the principles of applied behavioral analysis with more developmental and relationship-based approaches that are likely to be appealing to a wide community of professionals as well as being more user-friendly for most families. The design was a randomized clinical trial, with a manualized treatment, properly trained therapists, and fidelity of treatment delivery was measured throughout. In addition, outcome was measured with blind assessors over two time points using the most up-to-date, standardized measures; and minimal attrition after 2 years of intervention was registered.

Nevertheless, a number of issues sprung to mind when reading the manuscript that pertain to the internal and external validity of the study findings.

Despite a relatively small sample size, treatment effects reached statistical significance. However, relative to the control group, cognitive gains (2/3 of SD) in the experimental treatment were of moderate magnitude. Similarly, Vineland scores in the experimental group certainly indicated gains of skills as a function of age, but as the standardized scores did not change, there was no real catch up in developmental rates compared to typically developing children. Most gains appeared attributable to language improvements with other domains showing minimal gains and some declines.

The statistical analysis of the trial data was conducted on treatment completers only while it might have been useful to include the 3 dropouts from the control group, using the classical approach of the last observation carried forward (LOCF). Despite remaining low, dropping-out occurred differentially across treatment modalities, and the relatively small sample size and magnitude of treatment effects may render the findings sensitive to this analysis. As a result, differences between the two interventions may have been either attenuated or inflated, depending on the clinical status of the drop-outs when exiting the study.

Next, with the exception of those obtained in cognitive testing sessions, direct observational measures of the children were limited, especially to evaluate outcome in less structured settings. Thus, most of the outcome measures relied on parental reports and, even though assessors were blind to treatment status, parents could not blind to group membership. More independent and naturalistic assessments of participants’ behaviors would have strengthened the ecological validity of the study.

Intriguingly, gains reported in the experimental group were mostly achieved during the first year of treatment while subsequent improvements in the second year of treatment were much less impressive. It may be that there is a critical age at which treatment effects are maximal, in which case treatment response in the EDSM group should be larger in those participants who started treatment earlier. Alternatively, true time differences in treatment administration may have existed, either because treatment goals and methods were more efficacious in younger children beginning the treatment or because treatment intensity decayed over time. Or, it may be that the chosen outcome measures are age-dependent in their sensitivity for capturing change. Finally, sleeper or lagged treatment effects might be considered, with an opportunity for treatment gains to increase later should a post-treatment natural follow-up become available. The latter possibility is perhaps illustrated in the actual first 2 years of the trial where cognitive/language scores increased sharply during the first year, with adaptive behavior improvements occurring only in the subsequent year. These competing hypotheses can probably be empirically tested although the sample size will limit this possibility.

Treatment subjects were randomized between the two groups, but there was a noticeable difference in the intensity of treatment received in both groups. Adding up all hours of weekly treatment for each modality, the ESDM group received weekly 36.7 hours of treatment as compared to 18.5 hours for the controls. Therefore, one must question whether or not the more pronounced improvements in the experimental group reflect specific treatment effects of the ESDM as opposed to more general, non-specific, effects due to uncontrolled differences in treatment time or intensity.

Further examination of the variance in the outcome measures (Vineland scale, Mullen Scales of Early Learning) in the Early Start Denver Model group shows that, despite equal variances between groups at baseline, the variance was much larger in the experimental group at 1-year follow up for the MSEL, and at 2-year for both measures. The much larger heterogeneity observed in the experimental group suggests that responders and non- responders within the experimental group should be further compared in order to identify individual-level factors which might predict treatment response or mediate treatment effects.

The final comment concerns the customary limitations to the external validity of studies of this nature. As is usually the case in tightly controlled randomized trials, children with some co-occurring medical conditions (epilepsy) or identified genetic anomalies (fragile X) were excluded in an attempt to increase sample homogeneity, the assumption being here that treatment effects maybe otherwise masked by noise due to heterogeneity. However, as rapidly accumulating findings of new and different genes involved in the etiology of autism tell us, the “idiopathic” group of children with autism is unlikely to represent a pure, etiologically homogeneous, group. Therefore, the classical exclusion rules for studies of autism, especially for treatment studies, should be reconsidered soon in order to enlarge the universe to which results can be generalized and to avoid leaving a rapidly growing proportion of children with ASD orphan from evidence-based knowledge. Moreover, including children with a wider range of genetic abnormalities may prove necessary in future treatment studies to tease apart the aforementioned variability in treatment response, provided that genetic information is brought in the design and analysis of randomized clinical trials.

All studies have points that require further examination and other important manuscripts will undoubtedly flow from this critical study. Needless to say, the study provides important and immediately clinically relevant information, particularly at a time when universal screening for ASD in 18-month-olds has been proposed.

Dr. Eric Fombonne Canada Research Chair in Child Psychiatry McGill University Montreal, January 25th 2010

Conflict of Interest:

None declared