The Effectiveness of Interventions for Adults with ASD: A Meta-Analysis

Friday, May 15, 2015: 5:30 PM-7:00 PM
Imperial Ballroom (Grand America Hotel)
T. M. Belkin1, A. D. Rodgers2 and J. H. McGrew3, (1)Indiana University- Purdue University Indianapolis, Indianapolis, IN, (2)Department of Educational, School, and Counseling Psychology, University of Kentucky, Lexington, KY, (3)Clinical Psychology, Indiana University- Purdue University Indianapolis, Indianapolis, IN

Most intervention research for individuals with Autism Spectrum Disorder (ASD) has focused on children. There are relatively few studies of interventions for adults with ASD, and even fewer that utilize group designs (Shattuck et al., 2012; Howlin & Moss, 2012). 


A meta-analytic framework was applied to synthesize the literature from group-based interventions for adults with ASD. Both between-groups design and pre-post design studies were included.


Computer-based database and ancestry searches, and manual searches of peer-reviewed articles and dissertations were used to identify studies meeting inclusion criteria. To be included studies had to: (1) use a group design to evaluate the outcomes of a treatment or intervention for adults (≥ 16 years, and/or mean age ≥ 18 years) with ASD, (2) report original data, (3) be published in English, and (4) include five or more participants in their experimental groups. We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines in conducting and reporting analyses (PRISMA guidelines; see Moher et al., 2009). Studies were coded for outcome effects, demographic variables (e.g., mean age, ethnicity), categorical moderators (e.g., intervention type, outcome category) and continuous moderators (e.g., duration, intensity, and length of sessions). A second rater coded twenty percent of the studies (k = 8). Using a random effects model, we calculated standardized mean difference effect sizes for studies reporting between-group data, and standardized mean gain effect sizes for studies reporting pre-post data. 

Sampling bias was assessed using fail-safe N analyses (see Rosenthal, 1979) and Egger’s test of the symmetry of funnel plots (see Egger et al., 1997). The trim and fill procedure (see Duval & Tweedie, 2000) was used to account for areas of bias in the data.  Effect size heterogeneity was tested using both the Q statistic and I2


We identified 39 group design intervention studies including data from 1,086 participants (Intervention group: = 757; Control group: n = 329).  The mean sample size across studies was 27.85 (SD = 24.16; = 39) and the mean age of study participants was 25.70 (SD = 6.20; = 37). Most participants were Caucasian (= 83.44%, SD = 16.22; k =9) and male (M = 80.35%, SD = 13.46%; k = 39). 

The average effect sizes were medium both for studies reporting between-group data (= .74, k = 21) and for studies reporting pre-post data (= .60, = 34). The studies reporting pre-post data were heterogeneous, as shown by an asymmetrical funnel plot; however, none of the moderators explained the variation in study effect sizes.  


Interventions are moderately effective in improving outcomes for adults with ASD.  The homogeneity across the controlled studies (= 21) and non-significant moderation across the studies reporting pre-post data (k= 34), suggests that interventions for adults with ASD, in general, are effective, and that any differences observed were not due to differences in the tested moderators (e.g., type of outcome being assessed, type of intervention, intensity of treatment).