Predicting early intervention outcomes in autism via individual participant data mega-analysis.

Abstract

Background: Autism early intervention meta-analyses have yielded important insights into questions such as 'what works' and 'for what' outcomes. However, because these studies primarily rely on study-level summary statistics, they may be limited with regard to more individualized insights. Mega-analyses utilizing individual participant data (IPD-MA) may be useful for these more individualized questions.

Methods: We conducted an IPD-MA on the Autism Early Intervention Research (AEIR) consortium dataset, comprising n = 582 autistic children (n = 121 female) across 11 datasets collected in clinical and community settings in the USA, Switzerland, Italy, Israel, and Australia. Children received between 3 and 27 months of early intervention (age at start 13-60 months) of varying intensity. Of these datasets, one originates from a randomized controlled trial (RCT). Another 7 datasets are paired up into 5 separate controlled-group design studies, while the remaining 3 datasets come from uncontrolled pre-post design studies. Early Start Denver Model (ESDM; n = 281, 62 female) was compared against non-ESDM treatment-as-usual/community approaches (n = 301, 59 female) (e.g., speech and occupational therapy, applied behavioral analysis, pivotal response training). Outcome variables were Mullen Scales of Early Learning (MSEL) age-equivalent scores, Vineland Adaptive Behavior Scales (VABS) standardized scores, and Autism Diagnostic Observation Schedule (ADOS) calibrated severity scores.

Results: Predictors such as sex and cumulative intervention intensity were largely not associated with change in outcomes. Age at intervention start and pre-intervention developmental quotient (DQ) were strong moderators across all outcomes. Earlier age at intervention start predicted more positive outcomes, while higher pre-intervention DQ predicted accelerated growth on VABS motor and MSEL outcome measures. Interventions were similar in their effects on MSEL and VABS outcomes. However, for ADOS outcomes, ESDM resulted in significantly declining trajectories over time, with the sharpest decline for high pre-intervention DQ individuals. In contrast, ADOS trajectories in non-ESDM diverge in opposite directionalities (i.e. increases or decreases) depending on individual's pre-intervention DQ.

Limitations: Data for the IDP-MA was contributed on a voluntary basis and only 1 study was contributed from an RCT. Therefore, the findings may be prone to self-selection bias that may limit generalizability. Inferences from the intervention type comparison (ESDM versus non-ESDM) may be limited due to the heterogeneous approaches lumped together in the non-ESDM category. Larger sample sizes are required for finer-grained comparisons of different intervention types.

Conclusions: Age at intervention start and pre-intervention DQ are key individualized predictors and the latter can interact with intervention type to moderate early intervention response.