Parsing early emerging heterogeneity related to autism spectrum disorder

早期识别和干预自闭症谱系障碍（ASD）的主要障碍是我们的局限性

A major impediment to early identification and intervention for autism spectrum disorder (ASD) is our limited understanding of how different children present signs as toddlers, including what risk symptoms coincide across multiple dimensions to predict outcome. Our objectives are to quantify behavioral and brain connectivity based subtypes of risk that model the variability of ASD symptom expression in a community sample of toddlers. We will then test the predictive validity of this approach in the same cohort of children at three years of age in order to identify risk profiles that differentially predict later cognitive, behavioral, and clinical features. First, we will implement two unsupervised data-driven computational approaches in a community sample of 3000 children between 18-24 months old in order to characterize clusters of risk profiles. We hypothesize that each approach will identify a proportion of high-risk individuals consistent with epidemiological estimates of ASD and associated developmental disabilities (e.g., language or global DD). Based on our preliminary data, we anticipate that ~300 children will be identified by these data-driven risk-profiling methods. We also hypothesize that distinct patterns of structural and functional connectivity will distinguish groups of at-risk children and that these groups will differ from low-risk children. All children will be scanned with the same brain imaging sequences and procedures implemented in the Baby Connectome Project and will be compared to data from 100 low-risk children from that project. Our neuroimaging sample of 300 children will be reassessed at age three with direct clinical assessment using gold-standard diagnostic instruments as well as parent report. This will allow us to validate the risk profiling approach implemented at 18-24 months, to compare with a current screening approach, and to refine the risk profiling approach with supervised training of prediction algorithms that incorporates behavioral/clinical outcome data. We expect this method for risk stratification/subtyping to better model the heterogeneity inherent to the early at-risk and resilient phenotypes, which will subsequently improve early identification/diagnosis efforts. These outcomes will have translational impact because improved methods for early identification in ASD are necessary for the successful development of efficacious, personalized early interventions.