Age-Binned Normalization of Vinelandtm-II Increases Variability in Standard Scores: Implication for Clinical Trials in ASD.

Background:  Adaptive behavior is the performance of daily activities required for personal and social sufficiency. Autism Spectrum Disorder (ASD) patients’ adaptive behavior is impaired, even in patients with normal or above normal IQ levels. The most commonly used instrument to measure adaptive behavior is the Vineland Adaptive Behavior Scales, second edition (Vineland^TM-II) which has been extensively used in ASD research and in clinical practice. More recently Vineland^TM-II has been used in clinical trials which aim to measure changes in adaptive behaviors over time. A given patient’s raw score on the Vineland^TM-II scale must be adjusted for the score expected by the corresponding healthy control group of the same age. Importantly, the Vineland^TM-II manual provides normalization tables for these conversions for different age ranges (i.e., age ‘bins’) for the Composite score as well as for the Socialization, Communication, and Daily Living Skills domains scores. Thus, continuous raw scores are mapped onto discretized expected scores defined by categories of age ranges. Since both age and behavioral development are continuous variables, it is expected that this form of normalization will induce a certain amount of error, especially when a given patient’s age deviates from the midpoint of the age bin. Moreover, additional error may be induced in change scores from clinical trials and longitudinal studies when patients move to an older age bin during the course of the study.

Objectives: To estimate the impact of the age binning on simulated clinical trial Vineland^TM-II change data and to develop a continuous normalization function for the age normalization.

Methods:  The means of the 9 Vineland^TM-II raw-scores were extracted from the manual and interpolated with smooth splines to derive continuous estimates of the developmental trajectory. Then 1000 “average” study subjects (i.e. lying on the mean of the distribution) were simulated with an enrolment age randomly selected between 5 and 7 years. The standard domain scores were estimated for enrolment and 12 or 24 week assessments and the variances in the differences were estimated. By definition the difference should be 0 (i.e. effect of treatment was modeled) but as a result of age quantization, variance was induced that was set into relation to the variance observed in several studies that employed Vineland^TM-II.

Results: Simulations of 12-week long Vineland^TM-II clinical trial data with subjects aged 5 to 7 years showed that age binning increased variability of domain standard change scores by up to 2 units. Since the naturally occurring variation in domain change scores is as low as 6 standard scores in some clinical trials, this increase in variability corresponds to up to 12.5% of the variance. As expected, the continuous normalization function generated zero increase in change score variability in the simulated clinical trial data.

Conclusions: The use of Vineland^TM-II in clinical trials should take into consideration the increased variability introduced by the age binned normalization. Larger sample sizes should be considered to compensate for this higher variability. Alternatively, validation of continuous age normalization will improve the Vineland^TM-II use in clinical trials and observational studies.