The study was a secondary analysis of cross-sectional initial-visit data from the publicly available Get SET Early dataset. After complete-case analysis, it included 1,502 children diagnosed with autism spectrum disorder and 570 typically developing children. The autism group included 1,171 boys and 331 girls and had a mean age of 26.2 months; the typically developing group included 332 boys and 238 girls and had a mean age of 25.2 months. In the original study, legal guardians provided informed consent and the University of California, San Diego ethics committee approved the protocol.

The researchers constructed Gaussian graphical models from ten variables measured with three standardized assessments. The Autism Diagnostic Observation Schedule supplied social affect and restricted and repetitive behaviors. The Vineland Adaptive Behavior Scales contributed communication, socialization, daily living skills and motor skills, while the Mullen Scales of Early Learning supplied fine motor, visual reception, receptive language and expressive language. The models were estimated with EBICglasso regularization, and the researchers examined network topology, node centrality, descriptive sex-specific patterns and conditional-dependence structures.

The two groups differed in global strength and in their prominent estimated edges. In the typically developing network, communication had strong associations with expressive language and daily living skills, with edge weights of 0.31 and 0.30. In the autism network, the largest edges linked fine motor with visual reception at 0.45 and receptive with expressive language at 0.37. The negative association between fine motor and restricted and repetitive behaviors was also estimated at −0.12 in the autism network, compared with −0.01 in the typically developing network. These values describe conditional associations and do not demonstrate causal relationships.

Communication had the highest strength centrality in the typically developing network, whereas receptive language had the highest strength in the autism network. A Network Comparison Test based on 1,000 permutations found greater global strength in the autism network, with a difference of 1.79 and p<0.001. Global strength is the sum of absolute edge weights and is not equivalent to network density. Exploratory bridge-centrality calculations identified receptive language and communication as the leading cross-domain connectors in the autism network, with bridge-strength values of 0.661 and 0.655. Because these raw statistics were not tested with a normalized inferential comparison, they do not establish reduced inter-domain integration in autism.

Case-dropping bootstrap analyses with 1,000 resamples indicated good stability for the autism network’s strength, closeness and betweenness indices, each with a CS coefficient of 0.75. For the typically developing group, the corresponding coefficients were 0.672 for strength and 0.595 for closeness. These results concern stability within the submitted EBICglasso specification and resampling procedure; the study did not examine whether the network remained similarly stable under alternative estimators, tuning parameters or node sets.

The researchers also estimated directed acyclic graphs to summarize conditional-dependence structures. In the typically developing model, communication was connected to daily living skills, socialization, receptive language and expressive language. In the autism model, social affect and restricted and repetitive behaviors were connected to receptive language. The authors did not interpret these orientations as temporal order, developmental cascades or causal effects. With cross-sectional data, statistically equivalent graphs can represent the same distribution, so arrow directions cannot resolve causal or developmental mechanisms.

Exploratory comparisons by sex found that none of the ten variables differed between boys and girls in the autism group after Benjamini-Hochberg false-discovery-rate correction; all q values were at least 0.228. Eight variables differed in the typically developing group after correction, with girls scoring higher on communication, socialization, daily living skills, visual reception, receptive language and expressive language, and boys scoring higher on social affect and restricted and repetitive behaviors. Network comparisons found higher global strength in autism among both boys and girls, with differences of 2.03 and 1.56, respectively, both with p<0.001. The much smaller female autism subgroup, 331 compared with 1,171 males, and the absence of sample-size-balanced sensitivity analyses limit these exploratory findings.

Several limitations constrain the proposed model. Structural vulnerability was an interpretive network descriptor rather than a validated scalar index, and no normalized formal comparison of inter-domain integration was performed. Edge-level Network Comparison Test p values were unadjusted, age was not included as a covariate, and unmeasured clinical, socioeconomic and contextual characteristics may have influenced the estimates. The selected nodes did not include executive functioning, sensory processing or play skills, which could affect the estimated topology and centrality ordering. The model has also not been tested in an independent autism cohort. The authors therefore characterize receptive language and communication as candidate central and bridging nodes and state that longitudinal, covariate-adjusted, externally replicated and sensitivity-tested studies are required before developmental, clinical or intervention implications can be established.

Subject of Research: Cross-sectional symptom-ability network analysis of a provisional autism structural vulnerability model in preschool children

Article Title: A structural vulnerability model of autism derived from symptom-ability network analysis

Article References: Li, H.-M., & Chen, Z. (2026). A structural vulnerability model of autism derived from symptom-ability network analysis. Frontiers in Psychiatry, 17, Article 1837552. https://doi.org/10.3389/fpsyt.2026.1837552

Image Credits: AI Generated

DOI: 10.3389/fpsyt.2026.1837552

Keywords: autism spectrum disorder, network analysis, receptive language, adaptive communication, preschool children, cross-sectional study, centrality, conditional dependence, sex differences, hypothesis-generating research, structural, vulnerability