Testing autism interventions: trials and tribulations.

In The Lancet today, Jonathan Green and colleagues report results from a multisite randomised trial in children with autism. The investigators compared a parent-training technique that targeted enhancement of the child’s social-communication skills (two of the three core defi cits in autism) with treatment as usual. The primary outcome was the social-communication score from the Autism Diagnostic Observation Schedule (ADOS), a widely used diagnostic tool. Secondary outcomes included parent–child interaction, child language, social communication, and measures of adaptive functioning. Today’s results showed no eff ect of the intervention on the primary outcome and positive eff ects on some but not all secondary measures, including parent report of language and communication and direct observation of parent–child interaction. This study furthers the fi eld by setting a new bar for the minimum standards of rigorous methodology needed in trials that have potentially far-reaching service and policy implications. Strengths include a large sample size, multisite randomised design with masked assessors, balance across treatment groups, a manualised treatment approach (including standards for fi delity and interrater reliability), and outcome measures that directly relate to treatment focus. Thus, in a fi eld in which minimum study standards have made it diffi cult to even look for literature to answer what works for autism, this study is an achievement. At the same time, today’s study exemplifi es the complexity of attempting to detect change in samples of young children with such a heterogeneous condition. There are very few positive published trials in autism, for behavioural interventions, traditional pharmacotherapy, or complementary/alternative therapies. Is this due to non-effi cacious treatments, lack of sensitive outcome measures, or heterogeneity of autism—or perhaps all three? The lack of eff ects on ADOS social-communication scores with this parent-training intervention dampens optimism about its ability to exert clinically signifi cant change in core symptoms of autism. However, Green and colleagues’ use of a theoretical model that used both proximal and distal eff ect analyses is especially important. Unlike in many other trials, symptoms of autism were appropriately used in today’s study as the distal treatment eff ect (as regards the primary outcome) with proximal variables theoretically linked to the treatment as secondary outcomes. In this case, parent-training intervention lent itself easily to the use of proximal parent–child interaction variables, to explore the mechanism of the intended treatment. Development of appropriate models, starting with direct treatment targets and following downstream eff ects, might be needed for truly rigorous behavioural trials to eventually detect mechanisms of action. In fact, the eff ect of attenuation in today’s trial (with more eff ect seen in proximal than distal outcomes) allowed the investigators to begin to distinguish active crucial components of the treatment. Investigation of these components can provide data for the ongoing debate about the importance of the ingredients in a given intervention. Some of these multiple components include: implementer (parent vs therapist training), setting (individual vs group and home vs clinic), style (discrete trial vs play-based or relationship-based), and dose (total intervention time vs intensity per time period). Head-to-head comparisons, in which all but one of these variables is controlled, might be needed to answer questions about which ingredients are most active and effi cacious. Another issue is the dearth of outcome measures that are sensitive to change in autism symptomatology,