# Investigating the ADOS-2 calibrated severity score: insights from the ELENA cohort

## TL;DR

This study compares two scoring methods of the ADOS-2 test for autism severity, finding that one method may reduce sensitivity to individual differences over time.

## Contribution

The study reveals that ADOS-CSS calibration reduces module-related variance but may distort measurement signal compared to raw scores.

## Key findings

- ADOS-RS explains more variance in symptom severity than ADOS-CSS.
- ADOS-CSS shows lower stability and weaker correlations with external measures like SRS-2.
- Calibration reduces module differences but may affect sensitivity to individual changes.

## Abstract

The Autism Diagnostic Observation Schedule (ADOS-2) is widely used to assess Autism Spectrum Disorder (ASD) symptom severity. To allow comparisons across modules and developmental levels, the Calibrated Severity Score (ADOS-CSS) was developed. However, concerns have been raised about potential changes in the measurement signal introduced by the calibration process, which may alter the signal captured by the ADOS-2 raw scores (ADOS-RS).

Using longitudinal data from the ELENA cohort (N = 145 children with ASD assessed at baseline and after 3 years), we examined the psychometric properties of the ADOS-CSS and ADOS-RS. The Social Responsiveness Scale (SRS-2) was used as an external reference. Bootstrap resampling was used to estimate means and 95% confidence intervals (CIs).

Variance explained by module, age, and IQ was higher for the ADOS-RS (R2 = 0.45, 95% CI: 0.38–0.52) than for the ADOS-CSS (R2 = 0.16, 95% CI: 0.10–0.24). Stability of ASD severity scores over time was lower for the ADOS-CSS (ρ = 0.34, 95% CI: 0.23–0.45) than for the ADOS-RS (ρ = 0.60, 95% CI: 0.51–0.68). Correlations with SRS-2 scores were consistently lower for ADOS-CSS (ρ = 0.14, 95% CI: 0.06–0.22) than for ADOS-RS (ρ = 0.24, 95% CI: 0.17–0.32). Similar patterns were observed for change scores (Δ indicating score differences between baseline and follow-up): ΔSRS-2 correlated more strongly with ΔADOS-RS (ρ = 0.19, 95% CI: 0.09–0.28) than with ΔADOS-CSS (ρ = 0.13, 95% CI: 0.06–0.21). Differences between ADOS modules were reflected in SRS-2 and ADOS-RS, but markedly attenuated in ADOS-CSS, suggesting that calibration reduces module-related variance but also induces changes in the measurement signal that may affect sensitivity to individual differences.

While the ADOS-CSS facilitates standardized comparisons across modules, clinicians and researchers should be aware of potential distortions in the measurement signal associated with the calibration process.

NCT02625116.