Sex differences in diagnostic accuracy of ADOS-2 Module 4 in adults

Background

The Autism Diagnostic Observation Schedule, Second Edition (ADOS-2) Module 4 is a widely used standardized instrument for diagnosing autism spectrum disorder in verbally fluent adolescents and adults. Despite its broad clinical application, concerns have been raised regarding potential sex-related differences in diagnostic performance, particularly the possibility of underidentification of autism spectrum disorder in females. However, empirical evidence examining these differences in adult populations remains limited, and few studies have systematically evaluated diagnostic accuracy separately by sex, especially in clinically referred samples.

Objectives

The present study aimed to examine the diagnostic accuracy of the ADOS-2 Module 4 original and revised algorithms in adult males and females, and to identify optimal cut-off scores based on statistical and clinical criteria.

Methods

Receiver operating characteristic analyses were conducted separately for males (n = 398) and females (n = 368), using clinical diagnosis of autism spectrum disorder as the reference standard. Diagnostic accuracy was assessed using area under the curve, sensitivity, specificity, and the Youden index. Positive and negative likelihood ratios were calculated to estimate post-test diagnostic utility. Optimal cut-off scores were determined by maximizing the Youden index. Analyses were performed for both the original and revised ADOS-2 Module 4 algorithms. Missing data were excluded listwise, and confidence intervals were estimated under a nonparametric framework.

Results

Both algorithms demonstrated good diagnostic accuracy in males and females. Area under the curve values ranged from 0.830 to 0.839 (all p < 0.001), with no statistically significant differences between sexes or between algorithms. The optimal cut-off identified by the Youden index corresponded to a score of 8 or higher across all groups. At this threshold, sensitivity ranged from 0.69 to 0.79 and specificity from 0.78 to 0.86. Positive likelihood ratios ranged from 3.51 to 5.04, indicating moderate diagnostic utility for confirming autism spectrum disorder, while negative likelihood ratios ranged from 0.27 to 0.36, reflecting limited ability to rule out the condition. Across both algorithms, a consistent pattern of sex-related differences was observed: females showed lower sensitivity but higher specificity compared to males. This pattern indicates a greater likelihood of false negatives in females and relatively fewer false positives, suggesting differential test functioning rather than differences in overall accuracy.

Conclusions

The ADOS-2 Module 4 demonstrates comparable overall diagnostic accuracy in adult males and females when evaluated using global performance indices. However, differences in sensitivity and specificity suggest that the instrument may function differently across sexes, with reduced detection of autism spectrum disorder in females. Although a higher cut-off score optimizes statistical classification, it may reduce sensitivity, particularly in women, potentially contributing to underdiagnosis. The likelihood ratio findings further indicate that the instrument has moderate utility for confirming, but limited capacity for excluding, the diagnosis. These findings highlight the importance of considering sex-related factors in the clinical assessment of autism spectrum disorder in adults and suggest that reliance on a single cut-off score may not be equally appropriate across populations. Future research should examine whether integrating additional clinical information or adopting flexible diagnostic thresholds could improve detection without compromising specificity.

References

No references