Intensive behavioural interventions based on applied behaviour analysis for young children with autism: An international collaborative individual participant data meta-analysis.

All 10 studies that supplied individual participant data were non-randomised and rated at serious risk of bias in at least one domain. Allocation was often by parental preference, location or pre-determined assignment, and outcome assessors knew which intervention children received in 9 studies, so measured differences may be overestimated. No study protocols were obtained, so the review could not check whether analyses followed pre-specified methods. Outcomes were not collected consistently; autism symptom severity was available from only 3 studies and language was measured with different tools that could not be combined. IQ was measured with different scales across studies, and the review assumed the scores were comparable, which weakens confidence in IQ findings. Most follow-up was only 1 or 2 years, with very limited data beyond 2 years and only one study with 7-year follow-up. Adverse effects, harms, family well-being and quality of life were not systematically collected. The review could not examine most planned child characteristics as moderators, and subgroup analyses had very wide confidence intervals. Sensitivity analyses including aggregate data from non-IPD studies produced larger effects, driven mainly by one study, so the overall result is unstable. The studies were conducted over more than 20 years, so usual care and educational contexts may differ from current practice.

The work was funded by the NIHR HTA Programme. No other funding or author conflicts are reported in the supplied text.

Do not read the 14-point IQ and 7-point adaptive behaviour differences as proof that intensive ABA-based intervention works. All included studies were non-randomised, often allocated by parental preference or location, and outcome assessors usually knew the intervention, so these effects may be overestimated.