Heterogeneity in Motor Dysfunction and Intervention Response in Preschool Children with Autism Spectrum Disorder: A Study Based on Multidimensional Behavioral Assessment and Growth Mixture Modeling

Objective ​ This study aimed to systematically investigate the heterogeneity of motor dysfunction in preschool children with Autism Spectrum Disorder (ASD) using multidimensional behavioral assessment tools combined with advanced statistical methods, including Latent Class Analysis (LCA), Multilevel Growth Modeling (MLM), Latent Growth Mixture Modeling (LGMM), moderated mediation analysis, network analysis, and machine learning. It further sought to examine the effectiveness of a data-driven personalized intervention and explore the underlying mechanisms of intervention response. Methods ​ The study consisted of two phases: a cross-sectional survey (n = 125) and a randomized controlled trial (RCT, n = 80). Baseline assessments utilized a self-developed motor function scale, the fine motor subtest of the Peabody Developmental Motor Scales-Second Edition (PDMS-2), and behavioral coding of social motivation. LCA was employed to identify behavioral subtypes. In the RCT, the experimental group (n = 40) received a 12-week personalized motor intervention tailored to LCA subtypes, while the control group (n = 40) received routine care. MLM was used to test intervention effects and moderation effects, LGMM to identify heterogeneous trajectories, and moderated mediation analysis and Random Forest algorithm to explore mechanisms and predictors. Results ​ LCA identified three distinct subtypes of motor dysfunction: "Generalized Deficit" (28.0%), "Social-Fine Motor Deficit" (45.6%), and "Mild Deficit" (26.4%). MLM revealed a significant time × group × subtype triple interaction (p < .001), with the personalized intervention showing the largest effect size for the "Generalized Deficit" subtype. LGMM identified two trajectories within the experimental group: "Rapid Responders" (65%) and "Slow Responders" (35%). Moderated mediation analysis indicated that the intervention's effect was partially mediated by improvements in social motivation, a pathway particularly significant for the "Generalized Deficit" and "Social-Fine Motor Deficit" subtypes. Network analysis identified "Social Participation" as the most central node. The Random Forest model confirmed "Baseline Social Motivation" as the top predictor of intervention response type. Conclusion ​ This research establishes a comprehensive "subtyping-intervention-prediction-mechanism" framework by integrating multidimensional behavioral assessment and advanced statistical modeling. It confirms significant heterogeneity in motor dysfunction among preschool children with ASD, validates the effectiveness of a personalized intervention based on data-driven subtyping, and highlights the core role of "social motivation" in the mechanism and prediction of intervention response. This provides robust methodological support and empirical evidence for achieving precision rehabilitation for ASD at the behavioral level.