A new psychometric validation study has established calibrated scoring thresholds and likelihood-ratio decision bands for the Italian version of the Ritvo Autism Asperger Diagnostic Scale–Revised (RAADS-R), offering a more precise tool for adult autism screening. While the RAADS-R is a widely utilized self-report instrument, previous international validations have often relied on polarized case-control designs that struggle to perform effectively in real-world clinical settings, where psychiatric comorbidities and heterogeneous patient flows frequently lead to sub-optimal specificity. By conducting a rigorous case-control evaluation across two distinct Italian samples, researchers sought to refine the scale’s internal structure, determine discriminatory cut-offs, and develop a graded interpretation framework suitable for multidisciplinary assessment environments.
The study analyzed two samples comprising 1,460 total participants, balancing autistic adults with comparison community adults on age and gender. Sample 1 included 376 autistic adults who underwent comprehensive, blind multidisciplinary evaluations to confirm their diagnoses, alongside 376 matched non-autistic community adults. Sample 2 involved 354 adults with previously established autism diagnoses and 354 comparison adults. Researchers utilized several sophisticated statistical methods, including confirmatory factor analysis, exploratory structural equation modeling, and receiver operating characteristic (ROC) analysis, to test the reliability, factor structure, and measurement invariance of the Italian RAADS-R. Internal consistency was found to be excellent, with a McDonald’s omega of 0.98, and the data supported a hierarchical model that aligns with the DSM-5 diagnostic criteria.
A key finding was the scale’s predominantly unidimensional structure. Although the items can be mapped to seven DSM-5 sub-criteria, bifactor analyses demonstrated that these subscales contribute little reliable variance beyond the total score itself. Dominance analysis revealed that the top three domains—social-emotional reciprocity (A1), relationship understanding (A3), and interests and cognitive style (B3)—carry over 60 percent of the discriminative weight in the diagnostic models. Conversely, the repetitive behaviors (B1) subscale provided minimal discriminatory information. Consequently, the researchers conclude that the RAADS-R total score should serve as the primary metric for screening, while subscale scores are best interpreted as qualitative content descriptors rather than independent diagnostic indicators.
Discriminative accuracy in Sample 1 yielded an area under the ROC curve (AUC) of 0.91, identifying a candidate cut-off of 92, which provided a sensitivity of 0.89 and a specificity of 0.81. Sample 2, which included individuals with previously established diagnoses, produced a higher optimal cut-off of 104, reflecting a broader upward shift in scores that may be influenced by participant recruitment streams and the process of autistic self-recognition following diagnosis. The researchers emphasize that these cut-offs are preliminary and require external validation in consecutive clinical referral populations. Applying the standard international cut-off of 65 in these samples resulted in prohibitively low specificity, reinforcing the need for these context-specific calibrations.
To provide a more granular diagnostic interpretation, the team implemented a four-band likelihood-ratio framework. This system classifies scores into bands ranging from low likelihood (≤ 70) to high likelihood (≥ 137). These bands offer clinicians a graded approach to screening, although the researchers warn that they must be used with significant caution in clinical practice. Because the accuracy estimates were derived against community controls rather than consecutive psychiatric outpatients, the specificity of these bands is likely to be lower in settings where non-autistic patients present with high levels of psychiatric co-occurring traits. The bands provide a useful structure for interpreting scores, but the study stresses that no specific score range should be used to rule out an autism diagnosis when clinical suspicion exists.
Gender differences in reported autistic traits also emerged, particularly regarding sensory processing. In both samples, autistic women reported higher sensory scores than autistic men, and item-level invariance testing suggested that this reflects a phenotypic difference rather than basic measurement bias. However, the study noted that this observation requires cautious interpretation, as it could potentially relate to differences in how individuals perceive or report their sensory experiences. Autistic women also scored higher than autistic men on the social-communicative composite, while no significant group-by-gender interaction was observed for the cognitive-repetitive composite.
The study’s design carries certain limitations that necessitate careful clinical application. As a case-control psychometric validation, the estimates for sensitivity and specificity are prone to inflation compared to what would be expected in a consecutive clinical screening flow. Furthermore, the comparison groups were self-declared non-autistic community adults, who may not fully represent the diagnostic complexity of individuals presenting to psychiatric services with comorbid conditions like anxiety, ADHD, or mood disorders. The researchers also noted that intellectual functioning was not explicitly controlled for, and the RAADS-R is only intended for use in individuals with average to high intelligence and education, as is standard for this instrument.
Ultimately, the Italian RAADS-R appears to be a psychometrically sound tool for clinical assistance when utilized within a multidisciplinary framework. The study highlights that the tool’s primary value lies in its reliable total score and its ability to act as a structured screener that supports, rather than replaces, the clinical interview and developmental history. By providing locally calibrated thresholds, the study offers a clearer path toward more informed referrals in Italy. Future research needs to focus on longitudinal assessments and validation in psychiatric-referral samples to determine how the RAADS-R performs under the conditions of complex differential diagnosis.
The findings reinforce the importance of understanding the limitations of self-report screening tools in the context of adult ASD assessment. Adults with high levels of compensatory strategies or those who are in low-demand life environments may report lower trait levels, necessitating that clinicians remain attentive to the cognitive costs of social adaptation reported during interviews. Similarly, the study serves as a clinical advisory to avoid over-reliance on any single threshold. The successful deployment of the Italian RAADS-R, when combined with evidence-based clinical practice, can facilitate more timely diagnostic recognition for adults who have historically faced significant delays in obtaining specialist evaluation.
A detailed examination of discordant classifications at the candidate threshold of 92 revealed distinct demographic differences. Among the 43 autistic adults who scored below the cut-off, participants were older, achieved higher education, and were nearly four times more likely to be parents of an autistic child than those scoring above 92. These individuals displayed uniformly lower scores across all diagnostic subscales, suggesting that learned compensation, lower symptom severity, or different referral pathways may obscure self-reported traits. Conversely, the 72 comparison adults who scored at or above the threshold were younger, had completed fewer years of education, and reported a three-fold higher rate of personal psychiatric diagnoses compared to true-negative comparison peers. Crucially, these false-positive respondents endorsed substantially more recent, state-like symptoms rather than lifelong developmental traits, indicating that temporary psychiatric distress or transdiagnostic symptoms can elevate self-report scores.
The four likelihood-ratio decision bands provided concrete statistical metrics to guide interpretation across score intervals. Scores in Band A, falling at or below 70, produced an interval likelihood ratio of 0.06, reflecting a low likelihood of autism in this community-compared sample. Band B, spanning scores from 71 to 91, yielded an inconclusive likelihood ratio of 0.55. Band C, covering scores from 92 to 136, operated as a positive screen with an interval ratio of 2.50. Finally, Band D scores of 137 and above yielded an interval likelihood ratio of 10.15, encompassing 54 percent of autistic participants in the primary sample compared to 5 percent of non-autistic peers. However, bootstrap resampling across 5,000 iterations indicated that cut-off boundaries carried statistical uncertainty. While the threshold between bands A and B at 71 remained stable with a 95 percent confidence interval between 65 and 74, the upper boundary at 137 yielded a wide interval between 116 and 153. Moreover, the likelihood ratio confidence interval for Band D spanned from 6.93 to 17.18, suggesting that the data support an interval ratio of at least 7 rather than strictly 10.
Participant referral background markedly influenced total scores and optimal cut-offs. Within the primary clinical sample, 241 autistic adults had entered assessment with a pre-existing suspicion that was subsequently confirmed, averaging a score of 144.5. In contrast, 72 individuals entered without prior suspicion and were diagnosed during evaluation, scoring markedly lower with a mean of 113.6. Excluding this unsuspected group shifted the optimal Youden cut-off upward from 92 to 98. Multivariate analyses of covariance also documented a consistent age effect across diagnostic categories: older participants endorsed fewer autistic traits, amounting to an estimated decrease of approximately six total score points per decade of age.
Convergent validity analyses with the Autism-Spectrum Quotient further supported the scale’s construct validity while highlighting domain-specific nuances. In a subset of 646 participants who completed both questionnaires, total scores exhibited a strong correlation of 0.86, which reached 0.88 within the comparison group but narrowed to 0.53 among autistic participants due to score ceiling effects. Domain-specific comparisons demonstrated the strongest concordance between social dimensions, where the Autism-Spectrum Quotient social skills and communication subscales correlated strongly with the RAADS-R social-emotional reciprocity and relationship understanding domains, achieving coefficients between 0.73 and 0.80. Conversely, the imagination subscale exhibited the lowest cross-scale associations, with correlation values ranging from 0.17 to 0.40. These psychometric relationships, paired with the lower reliability observed for repetitive motor behaviors, reinforce that the self-report scale functions primarily through its unified assessment of social and cognitive autistic features.
Subject of Research: Psychometric validation and calibration of the Ritvo Autism Asperger Diagnostic Scale–Revised (RAADS-R) for Italian adult populations.
Article Title: Validation and Likelihood-Ratio Calibration of the Italian Ritvo Autism Asperger Diagnostic Scale–Revised (RAADS-R)
Article References: Tortora, C., Panzeri, M., Ronconi, L., Gasparini, A., Moscone, D., & Vagni, D. (2026). Validation and Likelihood-Ratio Calibration of the Italian Ritvo Autism Asperger Diagnostic Scale–Revised (RAADS-R). Journal of Autism and Developmental Disorders. https://doi.org/10.1007/s10803-026-07612-w
Image Credits: AI Generated
DOI: 10.1007/s10803-026-07612-w
Keywords: RAADS-R, Autism Spectrum Disorder, Diagnostic validation, Psychometrics, Adult assessment, Italian population, Likelihood-ratio calibration, Screening thresholds, Sensitivity and specificity, DSM-5 alignment, Clinical decision-making, Gender differences, Sensory processing, Multidisciplinary evaluation, Psychiatric comorbidity

