Assessing risk of bias in randomized controlled trials for autism spectrum disorder

Abstract

Aim: To determine construct validity and reliability indicators of the Cochrane risk of bias (RoB) tool in the context of randomized clinical trials (RCTs) for autism spectrum disorder (ASD). Methods: Confirmatory factor analysis was used to evaluate a unidimensional model consisting of 9 RoB categorical indicators evaluated across 94 RCTs addressing interventions for ASD. Results: Only five of the nine original RoB items returned good fit indices and so were retained in the analysis. Only one of this five had very high factor loadings. The remaining four indicators had more measurement error than common variance with the RoB latent factor. Together, the five indicators showed poor reliability (ω = 0.687; 95% CI: 0.613-0.761). Conclusion: Although the Cochrane model of RoB for ASD exhibited good fit indices, the majorities of the items have more residual variance than common variance and, therefore, did not adequately capture the RoB in ASD intervention trials.