Leveraging Machine Learning Techniques to Elucidate Risk for Callous-Unemotional Traits

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Callous-unemotional (CU) traits, defined by low empathy, guilt, and prosociality, predict very high risk for childhood disruptive behavior disorders (DBD) and adverse adult outcomes, including violence, psychopathy, and crime. Standard treatments for DBDs are not as effective for children with CU traits. To inform personalized treatments for DBDs, a better understanding is needed about the specific risk factors for CU traits beginning in early childhood. Prior studies are limited by focusing only risk factors within a single risk domain or at a single age point. Thus, based on extant literature, we do not know which risk factors for DBDs and CU traits matter the most nor at what age they matter the most, including the possibility that the most influential mechanisms are characterized by interactions and nonlinear associations across domains and ages. Moreover, while prior studies have begun to identify risk factors for CU traits within the Cognitive and Negative Valence Systems of the Research Domain Criteria (RDoC), there is a major knowledge gap and fewer available measures focusing on links between the Social Processes domain and CU traits. To address these knowledge gaps, the objectives of this R21 proposal are to: (1) Implement a newly-developed behavioral coding paradigm that assesses affiliation (e.g., verbal and physical displays of affection) and social communication (e.g., eye-gaze, engagement, synchrony) during parent-child interactions; (2) Employ automated methods to identify objective linguistic markers of these domains; and (3) Use machine learning (ML) approaches to identify the domain-specific and age-specific precision risk factors that best predict CU traits across early childhood and middle childhood. We achieve these objectives using existing data from the Durham Child Health and Development Study (DCHDS) (n=206), which includes extensive observational, biological, and questionnaire report data on a diverse sample of children and their families assessed 7 times during early childhood (18, 24, 30, and 36 months) and middle childhood (5, 6, and 7 years), with parent-report measures CU traits at 7-8 years old. We will test child-, parent- , and context-level risk factors for CU traits across different units of analysis (i.e., biological, report, observed) and across two developmental stages (early childhood and middle childhood). This proposal is innovative because it will leverage computational linguistic methods and a new observational coding paradigm to assess affiliation and social communication, which have vital transdiagnostic implications for understanding risk for mental illness. The current proposal will also open new horizons by identifying age-specific and domain-specific risk factors, fundamentally advancing knowledge of how CU traits develop. The proposed R21 research is significant because it improves our ability to assess individual differences within the RDoC Social Processes domain, and will identify domain-specific and age-specific risk factors for CU traits, fundamentally advancing our understanding of the development of CU traits and our future ability to develop personalized and age-specific treatments for CU traits.