Table 1 contains the sex distribution and demographic characteristics of the four subject groups.
Diagnostic Accuracy (Sensitivity and Specificity)
The total RAADS scores ranged from 44 to 227 in the ASD subjects and from 0 to 65 in the comparison groups. By ROC curve analysis, we determined that the best threshold for distinguishing between the two groups was a score of 65. Using this value, all 578 comparison subjects were correctly predicted with no “false positives” (specificity = 100%).
Six ASD subjects from three of the nine centers scored below the threshold (non-ASD predicted range). These “false negatives” yielded a sensitivity of 97% (see Table 2).
Table 3 shows the sensitivity, specificity and accuracy for the original four subscale domains and the four factors as well as the total RAADS-R score. As can be seen, the total RAADS-R score has the highest accuracy.
One-way ANOVA comparisons across all groups, mean RAADS-R scores, and statistical characteristics of subject groups are contained in Table 4. The means and statistical characteristics obtained after collapsing the four groups into two (all ASD subjects vs. all comparison subjects) are shown in Table 5. The mean RAADS-R scores of the ASD, the autistic, and the Asperger’s disorder groups (Tables 3 and 4) were significantly different from the means of the comparison groups (p < .0001). This was also true at each of the nine research centers. The original subscale domains, the factors and the total score are all significantly different in ASD versus non ASD.
An age and sex adjusted ANOVA of total RAADS-R scores and subscale scores, across diagnostic groups showed group differences. Sex was not significant as a main effect (p = .06) but sex-diagnosis was significant (p = .0043). The effect of diagnosis slightly varied by sex, however, diagnosis was the strongest effect. Age was significant as a main effect (p = .0035) and age-diagnosis effect was also significant (p < .05). In ASD subjects, an increase in age was positively correlated with an increase in mean RAADS-R scores. (Spearman correlation 0.37, p = .0003). However, age–sex–diagnosis effect was not significant (p = .1299). Repeated analyses after adjustment for sex and age between and among the subject groups showed significant differences between ASD and comparison groups (p < .0001).
Pair-Wise Comparison of Each Question
Post-hoc unequal variance t tests were performed for pair-wise comparison of each of the 80 questions. The ASD subjects had significantly greater scores on each of the individual RAADS-R items (N = 80) than either of the comparison groups (p < .0001 for all comparisons). Total RAADS-R mean scores between autistic and Asperger’s disorder subject groups were not significantly different.
All ASD subjects met the research diagnostic criteria of each center and had an ADOS module 4 score in the autism range (these were inclusion criteria). SRS-A scales were administered to 69 first-degree relatives of ASD subjects. The concordance rate was 95.59% (all but three scored in the “autism” range).
Analysis of the Subscales
Mean scores and Cronbach alpha correlation coefficients were computed for each of the four subscale domains described previously: circumscribed interests = .903, language = .789, sensory motor = .905, social relatedness = .923 (for the four participant groups’ mean scores of the subscales, see Table 6).
Factor Analysis was carried out using both varimax (orthogonal) and oblique (non-orthogonal) varimax rotation. The oblique rotational method was preferred because it assumes the underlying constructs to be interrelated. In addition, the oblique model had simpler factor structure that was easier to interpret.
Four factors were identified using the oblique rotation method. These factors had a simpler factor structure with much less cross loading compared to the orthogonal results, making for easier interpretation. These final oblique factors also had a higher correlation with the original subscales compared to the orthogonal factors. Factors I and IV correlated most strongly with the social construct (r = .80). Factor II correlated most strongly with circumscribed interests (r = .78), factor III with sensory motor (r = .88). The language construct was not picked up with the four-factor model. The variables determined below are based on factor scores.
Factor I: Social Relatedness, containing questions related to empathy, intimacy and social language accounted for 23.9% of the variance. Questions: 1, 6, 8, 11, 13, 14, 15, 16, 18, 31, 37, 38, 43, 48, 52, 53, 58, 62, 66, 68, 72, 74, 77; α = .93.
Factor II: Circumscribed Interests, also with questions relating to social blindness, accounted for 25.0% of the variance. Questions: 3, 4, 5, 7, 9, 12, 16, 17, 20, 22, 25, 27, 28, 32, 33, 39, 41, 44, 45, 49, 50, 56, 60, 64, 76, 78, 79, 80; α = .95.
Factor III: Sensory Motor accounted for 22.4% of the variance. Questions: 2, 10, 19, 24, 29, 34, 35, 36, 40, 46, 51, 54, 57, 59, 65, 70, 71, 73; α = .87.
Factor IV: Social Anxiety accounted for 18.9% of the variance. Questions: 21, 23, 26, 30, 42, 47, 55, 61, 63, 67, 69, 75; α = .89.
A confirmatory factor analysis using these four factors showed that the coefficient of variation of the factor loadings was 20, 18, 18, and 19%, respectively, implying similar loadings for each item within a given factor. A comparison of factor loadings (differential item function) by gender and by ASD vs. non-ASD showed no significant differences with the mean difference of 18% or less.
Fifteen ASD subjects and fifteen comparison subjects without another DSM-IV-TR diagnosis were retested at UCLA after a mean interval of 1 year (maximum interval 15 months, minimum interval 10 months) The initial RAADS-R mean score for the ASA subjects was 155, and when repeated was 149. The initial RAADS-R mean score for the comparison subjects was 20, and 21 when repeated (Table 7 contains the statistical analyses for test–retest data).
The total mean RAADS-R scores for the ASD subjects from each of the 9 centers were significantly different from the comparison groups (p < .0001) and there were significant differences between the universities (see Table 8).
Clinical Implications of RAADS-R Scores
A RAADS-R score of 65 or greater is consistent with a clinical diagnosis of ASD. A RAADS-R score of 64 or lower is not consistent with a diagnosis of ASD (sensitivity = 97%, specificity = 100%). It must be emphasized, however, that if a subject has a score of 64 or lower but clinical judgment indicates that ASD is present, the clinical judgment should take precedence. This is due to the many limitations of self-rating scales that will be elaborated upon in the discussion section.