Age-stratified individual-specific subspace of autism spectrum disorder based on common orthogonal basis extraction algorithm improves the accuracy of clinical symptoms prediction.

backgroundAutism spectrum disorder (ASD) is a group of highly heterogeneous neurodevelopmental disorders with onset in early childhood. Functional magnetic resonance imaging (fMRI) studies have revealed that ASD is related to altered functional connectivity (AFC), and that individual-specific change isolated from AFC with data-driven method of individuals with ASD has enhanced the predictive ability for behavioral symptoms. Although few studies have incorporated age as a factor, it is critically important for ASD, a neurodevelopmental disorder, as age influences the disorder's onset and progression.

methodsIn this study, we analyzed 437 participants (208 ASD, 229 typical development) from the Autism Brain Imaging Data Exchange, employed the common orthogonal basis extraction (COBE) algorithm to isolate age-stratified individual-specific features and examine their predictive abilities for clinical behaviors. A validation analysis was performed in an independent sample.

resultsWe found that the age-stratified, individual-specific features improved behavioral prediction. The most substantial improvement was observed in predicting social behavior among adolescents with ASD, which showed a peak increase of 133% (0.35 [var = 0.10] vs. 0.15 [0.15]) and an average increase of 41% compared to AFC. These findings were replicated in an independent validation dataset.

conclusionThe age-stratified individual-specific features demonstrate superior predictive power for clinical symptoms. This underscores the critical importance of incorporating both inter-individual variability and the developmental perspective into ASD biomarker exploration and targeted intervention research.