Effectiveness of Responsivity Intervention Strategies on Prelinguistic and Language Outcomes for Children with Autism Spectrum Disorder: A Systematic Review and Meta-Analysis of Group and Single Case Studies.

Twenty-five of the 33 randomised trials were rated at high overall risk of bias and only one was rated low, mostly because outcome measurement was at risk of correlated measurement error. Time in intervention could not be analysed for either set of studies, so this review says nothing about how much intervention is needed or whether more is better. The Egger's test suggested publication bias against small studies with negative results in both sets of studies, so the pooled numbers are likely to be optimistic. Almost all of the single-case studies measured behaviour only in situations close to the treatment context, and generalisation probes were usually too infrequent to meet quality standards. The authors say the single-case results should be read with caution for this reason. Many of the single-case studies used only three participants, so whether one child responded or not could change the overall judgement of whether the intervention worked. Forty-four single-case studies that would otherwise have been included were excluded because they failed quality standards, so the included set is not the whole literature. Some included papers shared the same participants, so the children counted across studies may overlap. Participants were young on average (mean 43.01 months in the trials and 54.36 months in the single-case studies), most studies were done in the USA, and studies not written in English were excluded, so the results may not transfer to other ages or settings.

The authors name the National Institute of Child Health and Human Development as the funder. No other funding source or conflict of interest is stated in the text provided.

Do not assume these gains carry over into new settings or to people who did not deliver the intervention. In the randomised trials the difference was clear only for outcomes measured in situations very like the treatment sessions, and the results that were free of correlated measurement error showed no significant effect.