Discovering novel predictors of minimally verbal outcomes in autism through computational modeling

通过计算模型发现自闭症最低限度语言结果的新预测因素

• 批准号: 10521901
• 负责人: NANCY CAROLINE BRADY
• 金额: $ 59.62万
• 依托单位: GEORGIA INSTITUTE OF TECHNOLOGY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-08-05 至 2027-05-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10521901
• 关键词: Address; Affect; Age; Age-Months; Attention; Behavior; Child; Child Development; Child Language; Classification; Communication; Communities; Comprehension; Computer Models; Data; Data Set; Development; Diagnosis; Dimensions; Disease model; Early Intervention; Failure; Foundations; Four-dimensional; Gestures; Individual; Intervention; Joints; Knowledge; Language; Lead; Life; Literature; Measurement; Methods; Modeling; Nature; Outcome; Pathway interactions; Pattern; Predisposing Factor; Research; Research Personnel; Risk; Risk Factors; Sampling; Science; Survival Analysis; Testing; Time; Training; Validation; alternative communication; autism spectrum disorder; autistic children; base; data-driven model; effective intervention; experience; high risk; innovation; joint attention; kindergarten; language outcome; markov model; minimal risk; model development; novel; predictive modeling; public health relevance; skill acquisition; skills; vocalization

The proposed project will create novel multidimensional models to characterize the prelinguistic developmental pathways leading to verbal and minimally verbal (MV) outcomes in children with autism spectrum disorder (ASD). Children with ASD experience significant delays in the development of prelinguistic communication (PLC) skills that are important indicators of progress along a path towards language. PLC skills include vocalizations, gestures, joint attention, and comprehension. As many as 30% of children with ASD remain MV, producing very few if any spoken words by the time they reach kindergarten. Significant gaps remain in our knowledge of early risk factors that predispose children to remain MV. In particular, further research is needed to identify specific inflection points that are indicative of risk for not progressing to spoken language. The proposed innovative modeling framework will assess whether transitions between specific prelinguistic stages and the timing of these transitions represent risk for MV outcomes. Crucially, the project will develop a novel method for quantifying a child’s risk of a MV outcome at age 5 given their PLC progressions at earlier points in development. Such information could guide and focus early intervention efforts, such as intensifying therapies at certain points in development and/or deciding when to introduce augmentative or alternative communication. The key innovation in this proposal is leveraging Continuous-Time Hidden Markov Models to delineate progressions of PLC across dimensions of attention, vocalizations, gestures, and comprehension from 18-36 months in children with ASD, and to identify unique multidimensional trajectories that predict which children remain MV at age 5. The proposal will test the hypothesis that children who are transitioning between PLC stages more slowly or following atypical patterns of progression are at higher risk for MV outcomes. Aim 1 will develop and validate state-based models of development of attention, vocalizations, gestures, and comprehension in a well-characterized sample of typically developing children and separately, children with ASD (Activity 1a), and then construct a multidimensional model that unifies the individual models to simultaneously examine progressions across the four dimensions (Activity 1b). Models will be first validated on a sample of 50 typically developing children observed every 3 months from 6 to 18 months of age, and then separately validated on a sample of 100 children with ASD observed every 3 months from diagnosis at 18-24 through 36 months of age. Aim 2 will utilize the ASD model from 1b to identify predictors for MV status at 5 years in children with ASD (Activity 2a) and apply a survival analysis approach to turn these predictors into a quantifiable risk score for MV outcome (Activity 2b). The models along with the underlying training data will be released to the research community, enabling ASD and developmental researchers with novel datasets to assess the extent to which their data is consistent with ours, thus advancing the field of developmental science and laying the foundations for targeted interventions.

拟议的项目将创建新的多维模型来描述前语言发展 导致自闭症谱系障碍儿童语言和最低限度语言（MV）结果的途径 （ASD）中指定的值。ASD儿童在语言前交流的发展方面经历了显着的延迟 (PLC)这些技能是语言发展道路上沿着进步的重要指标。PLC技能包括 发声、手势、共同注意和理解。多达30%的ASD儿童仍然是MV， 到了幼儿园时，即使能说出来的话也很少。我们的差距仍然很大 了解使儿童易于保持MV的早期风险因素。尤其需要进一步研究 以识别特定的拐点，这些拐点指示不发展为口语的风险。的 提出的创新建模框架将评估特定的前语言阶段之间的过渡是否 并且这些转变的时间代表MV结果的风险。最重要的是，该项目将开发一部小说， 一种量化儿童在5岁时MV结果风险的方法，考虑到他们在早期的PLC进展， 发展这些信息可以指导和集中早期干预工作，如强化治疗 在发展和/或决定何时引入辅助或替代交流的某些阶段。 该方案的主要创新之处在于利用连续时间隐马尔可夫模型来描述 PLC在注意力、发声、手势和理解方面的进展， 月，并确定独特的多维轨迹，预测哪些儿童 5岁时保持MV。该提案将测试一个假设，即在PLC之间过渡的儿童 分期更慢或遵循非典型进展模式的患者发生MV结局的风险更高。目标1将 开发和验证基于状态的注意力发展模型，发声，手势， 理解在一个典型的发展中的儿童和单独的，儿童的良好特征的样本， ASD（活动1a），然后构建一个多维模型，将各个模型统一起来， 同时检查四个维度的进展（活动1b）。模型将首先在 从6至18个月大，每3个月观察一次50名发育正常的儿童样本，然后 在100名ASD儿童的样本中进行了单独验证，从18-24岁诊断开始，每3个月观察一次 直到36个月大。目标2将利用1b中的ASD模型来识别5处MV状态的预测因素 年的ASD儿童（活动2a），并应用生存分析方法将这些预测因子转化为 MV结局的可量化风险评分（活动2b）。 这些模型沿着底层训练数据将发布给研究社区，从而实现ASD 和发展研究人员与新的数据集，以评估他们的数据是一致的程度， 这是我们的使命，从而推动发展科学领域的发展，为有针对性的干预措施奠定基础。