Mega-analytic support for Early Start Denver Model, age at intervention start, and pre-intervention developmental level as factors differentiating early intervention outcomes in autism

Abstract

Objective: Autism early intervention meta-analyses have provided initial answers to questions such as 'what types of interventions work' and 'for what outcomes'? However, we also want to know 'for whom' is early intervention most effective for? Mega-analysis can offer up complementary insights to meta-analyses regarding the 'what works' and 'for what', while also offering unique insights into the 'for whom' question.

Methods: Here we conduct a mega-analysis with linear mixed effect modeling on AEIR consortium early intervention datasets totaling n=645 children spanning several countries (e.g., USA, Switzerland, Italy, Israel, and Australia). Early Start Denver Model (ESDM) and other non-ESDM approaches (e.g., EIBI, NDBI, other community/treatment as usual approaches) was evaluated as contrasting intervention types. Models also evaluated intervention intensity, type, participant sex, age at intervention start, and pre-intervention developmental quotient. Subscales of Mullen Scales of Early Learning (MSEL), Vineland Adaptive Behavior Scales (VABS), and Autism Diagnostic Observation Schedule (ADOS) were utilized as outcome measures.

Results: Neither intervention intensity nor participant sex affected outcomes. ESDM showed faster growth in language and non-verbal cognition compared to non-ESDM intervention. Irrespective of intervention type, earlier intervention start was associated with increased MSEL and VABS scores and decreased ADOS severity. Growth trajectories on the MSEL also varied by pre-intervention developmental quotient, with higher quotients predicting faster growth irrespective of intervention type.

Conclusions: Age at intervention start and pre-intervention developmental quotient are important individualized factors that predict early intervention response. ESDM also impacts language, non-verbal cognition, and core autism features.

Figure 1:. The Autism Early Intervention Research (AEIR) consortium along with intervention and characteristics describing the combined AEIR dataset.

Panel A shows sites around the world where data from the AEIR consortium originates from. On the far left of Panel B is a plot that shows the proportion of males (blue) versus females (pink) broken down by the two intervention types (ESDM, left; non-ESDM, right). The plot on the mid-left shows age at intervention start density plots for the two intervention types (ESDM, purple; non-ESDM, green), where age at intervention start is indicated in months on the x-axis. The plot on the mid-right shows the intervention intensity index (see Methods for a description of how this index is computed) for the two intervention types. The plot on the far right shows pre-intervention developmental quotients (DQ; x-axis) for the two intervention types.

Figure 2:. Comparison of ESDM versus non-ESDM intervention.

Panel A is a hypothetical figure showing an example of a significant intervention type*age interaction effect, whereby one intervention type (e.g., ESDM) shows faster growth over time than the other intervention type (e.g., non-ESDM). Panels B-D show spaghetti plots for ADOS (B), VABS (C) and MSEL (D) outcome measures. In all plots, ESDM is shown in purple, while non-ESDM is shown in green. The x-axis shows chronological age in months, while the y-axis shows the dependent variable for each measure (e.g., CSS scores for ADOS, standardized scores for VABS, age-equivalent scores for MSEL). Each individual child is shown with transparent lines, while the group-level trajectory along with 95% confidence bands are shown overlaid on top. The ~ symbol indicates intervention type*age interaction effects passing FDR q<0.1, while a star (*) indicates intervention type*age interaction effects that passing FDR q<0.05.

Figure 3:. Effect of age at intervention start and pre-intervention developmental quotient on Mullen (MSEL) and Vineland (VABS) scores.

Panel A and C show hypothesized examples of a main effect of age at intervention start (A) and an interaction between pre-intervention developmental quotient (Pre-Int DQ) and age (C). The coloring in all plots indicates continuous variation in each decile of the main predictor variable. For age at intervention start, the lowest deciles (lighter blue) are for individuals starting intervention at earlier ages. Increasing deciles (progressively darker blue) scale with progressively later age at intervention start. For Pre-Int DQ, the lowest deciles (darker red) are for individuals with the lowest Pre-Int DQ. Increasing deciles (progressively lighter red) scale with increasing levels of Pre-Int DQ. The plot in panel A shows a hypothetical main effect of age at intervention start, whereby there generally higher scores are observed for individuals that start intervention at the earliest ages. Note that there is no interaction with age, and this indicates a situation where the trajectories stay parallel as one increases from starting earlier to later. In contrast, Panel C shows a hypothetical interaction between Pre- Int DQ and age. In this situation, the slope of trajectories change as one goes from lower Pre- Int DQ (e.g., shallower slopes indicative of relatively slower growth) to higher Pre- Int DQ (e.g., steeper slopes indicative of relatively faster growth). Panels B and D show spaghetti plots of the actual data. Chronological age in months is plotted on the x-axis, while the outcome dependent variable is plotted on the y-axis (e.g., standardized scores for VABS, age-equivalent scores for MSEL). Individuals are shown with the transparent lines and decile trajectories are shown with thicker lines overlaid on top. Panels B and D shows the effects of age at intervention start (B) and Pre- Int DQ (D) on Mullen (MSEL, top) or Vineland (VABS, bottom). In panel B all outcome measures show a significant main effect of age at intervention start (e.g., indicated with the * for a significant effect passing FDR q<0.05), indicative of a stacking pattern with lighter blue deciles of individuals that started treatment earlier tending to have generally higher scores on the outcome measures. The Mullen Fine Motor also has a significant age at intervention start by age interaction (e.g., indicated by the star and brackets, ]*, passing FDR q<0.05). This effect can be explained by flatter slopes for individuals starting intervention earlier and steeper slopes for individuals that started later. Panel D shows Pre-Int DQ interaction with age for MSEL outcomes (e.g., ]*) and a main effect of Pre-Int DQ with no age interaction for VABS outcomes.

Figure 4:. Effects of age at intervention start and pre-intervention developmental quotient on autism calibrated symptom severity (ADOS CSS) scores.

The plots in this figure depict the effects of age at intervention start (A) or pre-intervention developmental quotient (Pre-Int DQ) (B) on ADOS CSS SA (left), CSS RRB (middle) and ADOS CSS Total scores (right). In these spaghetti plots, the x- axis shows chronological age in months, while the y-axis shows ADOS CSS scores from 0–10, with 0 being least severe and 10 being most severe. Individuals are shown with the transparent lines, while decile trajectories are shown with thicker lines overlaid on top. Interaction effects between the predictor variable and age are annotated with a star and bracket (e.g., ]*) or ~ and bracket (e.g., ]~). A tilde symbol (e.g., ~) represents an effect passing FDR q<0.1, while the star (*) represents a significant effect passing FDR q<0.05. The coloring in all plots indicates continuous variation across deciles of the predictor variable. For age at intervention start, the lowest deciles (lighter blue) are for individuals starting earliest, with increasing deciles (darker blue) scaling with later age at intervention start. For Pre-Int DQ, the lowest deciles (darker red) are for individuals with the lowest Pre-Int DQ and increasing deciles (lighter red) scale with increasing levels of Pre-Int DQ.