Comprehensive Meta-Analysis of Early Intervention Research for Young Children with ASD

Background: Early-intervention research for young children with ASD has been rapidly proliferating, partly owing to the assumption that intervention provided before school entry will have the largest impacts on later outcomes. However, there are marked differences between early intervention approaches, and there is currently no consensus on the ‘best’ strategies for supporting developmentally important outcomes for this group of children. Meta-analysis is a useful tool for sorting and synthesizing extant research, to gain a holistic understanding of research evidence.

Objectives: The purpose of this project is to generate summary effect sizes and tally quality indicators from all available studies that investigated the effect of early interventions on any outcome for young children (ages 0-8) with ASD.

Methods: To gather peer-reviewed literature and dissertations/theses, we searched nine online databases. The initial search yielded 12,933 records. In an effort to gather “grey literature”, or studies not published in a peer-reviewed journal, investigators who received federal grants to study autism were emailed and asked to provide unpublished data that would fit our inclusion criteria. A preliminary screen of abstracts was first completed using Abstrackr software. Studies that met the following inclusion criteria: (a) published in English, (b) published from 1970 - present, (c) group design that includes both an intervention and control group, (d) participants received a confirmed ASD diagnosis, and (e) participants were within the 0-8 year old range, went on for a full-text reading. Following screening, two coders extracted relevant information and effect sizes from each study. Discrepancies were resolved by consensus. We used robust variance estimation (RVE; Hedges, Tipton, & Johnson, 2010) to synthesize Hedge's g effect sizes within each intervention and outcome type. The RVE approach accounts for the nesting of multiple effect sizes within a single study sample, allowing us to take all available effect sizes from each study without violating independence assumptions.

Results: The search and screening process yielded 1,615 effect sizes gathered from 347 studies/databases. The RVE approach requires that at least five studies contribute to the generation of effect sizes. This criteria was met for six intervention categories; traditional behavioral (8 outcomes), naturalistic developmental behavioral (7 outcomes), developmental (2 outcomes), computer-based (1 outcome), sensory (1 outcomes), and TEACCH (2 outcomes). Significant summary effects ranged from 0.25 to 0.44. Coefficients, standard errors, and confidence intervals for each summary effect is listed in Table 1. However, a high percentage of behavioral studies were quasi-experimental in design, and all of the intervention types had high risk of bias for at least one study quality indicator.

Conclusions: In this comprehensive meta-analysis, we computed 21 summary effects across 6 intervention types. Results indicate at least some support for behavioral, NDBI, developmental, and computer-based interventions across several outcome types, and no support for any outcomes for sensory and TEACCH interventions. Future intervention research should include studies with randomized controlled designs, and low risk of bias across the full set of quality indicators. These findings have implications for best practices within the early intervention period.