The Minimal Clinically-Important Difference (MCID) on the Vineland-II: Analysis of Data from 3,400 Individuals with ASD

Background: Autism Spectrum Disorder (ASD) is associated with impairments in adaptive abilities in multiple domains including personal, social, and communication. Relative to age-matched peers, these impairments become increasingly pronounced across development, and are present regardless of comorbid intellectual disability. The Vineland Adaptive Behavior Scales, Second Edition (Vineland^TM-II) is the most commonly used instrument for quantifying these impairments but Minimal Clinically-Important Differences (MCIDs) on the Vineland^TM-II have never been established. Determining MCIDs for the Vineland^TM-II would facilitate the evaluation of new interventions for ASD, given the increasing use of the Vineland^TM-II in clinical trials.

Objectives: To generate anchor-based and distribution-based estimates of the MCID on the Vineland^TM-II adaptive behavior composite and standard scores for the socialization, communication and daily living subdomains.

Methods: We pooled data from several consortia/registries (EU-AIMS LEAP study, ABIDE-I, ABIDE-II, INFOR, Simons Simplex Collection) and clinical trials (Stanford, Yale, Roche). Quality control procedures were applied to the combined dataset (n>3,400 individuals), including the re-derivation of domain-level scores and outlier exclusion. The combined sample was stratified by age (children, adolescents and adults) and presence of comorbid intellectual disability (full-scale IQ < 70). Within each stratum, we adjusted for the fixed effects of age, gender, full-scale IQ, and the method of Vineland administration; we also adjusted for the random effect of dataset where possible. Two approaches were used to estimate the MCID: distribution-based methods and anchor-based methods. Distribution-based MCIDs include the standard error of the measurement, as well as one-fifth and one-half the covariate-adjusted standard deviation (both cross-sectionally [3,467 observations] and longitudinally [348 observations]). Anchor-based MCIDs include the slope of linear regression of CGI-S on Vineland score (630 observations), the slope of linear regression of CGI-I category on Vineland change (848 observations), the Vineland change score maximally differentiating minimal from no improvement on the CGI-I, the linking equation (Fayers & Hays, 2014), and equipercentile equating. Each MCID is reported separately, in addition to sample size-weighted averages for each MCID approach (anchor- vs. distribution-based).

Results: A Working Group was formed with representatives from Autism Speaks, Duke, King’s College London, Roche, Simons Foundation, Stanford, and Yale. The first outcome was a consensus Statistical Analysis Plan (SAP) describing the QC process and methods for MCID estimation. At the time of writing, estimates of the Vineland^TM-II MCID individuals fall within the range of 1.83 to 4.67 points for the Vineland composite score across strata (d-MCID range: 1.83-3.15; a-MCID range: 2.05-4.67), with a trend towards lower MCIDs in the younger and intellectually-disabled populations due to increased variance among adults. MCIDs will be further refined as our pooled sample size increases.

Conclusions:  Collaboration across academia, advocacy groups, and industry is critical to establishing a consensus on interpretation of clinical endpoints. Use of the Vineland^TM-II as an outcome instrument must take into account both the statistical and clinical significance of any observed change. Our ongoing analyses will inform the evaluation of new interventions for ASD by providing estimates of the Vineland™-II MCID across a wide range of ages and intellectual abilities.