Multi-level Modular Agent-based Modeling for the Study of Childhood Obesity

描述(由申请人提供)：我们建议开发和应用一种新的模块化的基于代理的建模方法，用于儿童肥胖的多水平研究，重点是饮食。这里设想的研究是基于大脑到社会系统(BTS)的方法，由首席研究员杜贝领导的跨学科团队在发育过程中研究饮食和肥胖预防，并包括与该提案最相关的每个领域的专家。在布鲁金斯学会的Co-I Hammond的领导下，我们建议在基于代理的建模工作的基础上建立多学科和多层次的工作，以开发基于代理的方法的新扩展--允许对每一级别的分析进行单独考虑的“模块”方法，但重要的是允许直接集成“模块”来研究多级别的反馈和交互。模块化的ABM将考察影响饮食行为和体重(BMI)的复杂生物/环境交互作用的五个水平的影响：(1)遗传学(多巴胺能基因系统DRD2和DRD4)；(2)神经生物学(多巴胺-纹状体和执行控制功能)；(3)心理倾向(约束/去抑制和对奖励的敏感性)；(4)家庭(孩子的依恋方式和母亲在怀孕/幼儿时期的焦虑/抑郁，环境逆境，食品安全和贫困)；(5)社会(母亲的社会规范和社会资本，家庭食物环境)。ABM生成的合成数据将与MAVAN队列(由共同-I Robert Levitan和Michael Meaney领导)现有的纵向经验数据进行比较，MAVAN队列是从怀孕和分娩时以纵向受试者内设计观察的母亲和她们的孩子的样本，以检查基因-环境相互作用和神经发育。这里提出的研究的具体目标是：(1)利用儿童早期数据(0-4岁)，为与儿童肥胖相关的五个层次--遗传、神经认知、心理、家庭和社会/环境--的每个层次构建单独的基于代理的“模块”模型，(2)整合ABM的多个模块，并探索各层次之间的反馈循环，(3)测试模块化ABM对5-7年关键过渡期的预测能力，以及(4)使用功能数据分析提供模块化ABM模型的进一步验证。完整的模型将使随后能够探索干预措施的潜在政策设计，以及预测已确定的趋势的影响。在该项目中开发的模块化ABM方法也将用于研究其他类似复杂的问题。简而言之，该项目将开发一种新的多水平方法(基于模块化代理的计算模型)，并将其应用于儿童肥胖的研究，以提高我们对儿童肥胖的多水平决定因素的理解，并帮助设计更有效的多水平干预措施，考虑影响儿童饮食和体力活动的生物、家庭、社区、社会文化、环境、政策和宏观经济因素的范围。 公共卫生相关性：该项目将开发一种新的多层次方法(基于模块化代理的计算建模；ABM)，并将其应用于儿童肥胖症的研究，以提高我们对儿童肥胖症的多层次决定因素的理解，重点是饮食。ABM生成的合成数据将与现有的纵向经验数据进行比较，这些数据来自一组社会经济地位较低的母亲及其子女，这些数据是从怀孕和出生时以纵向受试者内设计观察的，以检查基因-环境相互作用和神经发育。结果将有助于设计更有效的多层次干预措施，考虑影响儿童饮食和体重的生物、家庭、社区、社会文化、环境、政策和宏观经济因素的范围。

DESCRIPTION (provided by applicant): We propose to develop and apply a novel modular agent-based modeling approach for the multilevel study of childhood obesity, with a focus on eating. The research envisioned here is anchored in the brain-to-society system (BtS) approach to the study of eating and obesity prevention in development by a trans-disciplinary team lead by principal investigator Dube, and includes experts from each of the fields most relevant to this proposal. Under the leadership of Co-I Hammond from the Brookings Institution, we propose to build upon agent-based modeling work that is both multi-disciplinary and multi-level, to develop a novel extension of the agent-based methodology-a "modular" approach that will allow separate consideration of each level of analysis, but importantly will permit straightforward integration of the "modules" to study multi-level feedbacks and interactions. The modular ABM will examine five levels of influence expected to modulate the complex biology/environment interactions influencing eating behaviors and body weight (BMI): (1) genetics (dopaminergic gene systems DRD2 DRD4); (2) neurobiology (dopamine-striatal and executive control functions); (3) psychological predisposition (restraint/ disinhibition and sensitivity to reward); (4) family (child's attachment style and mother anxiety/depression during pregnancy/early childhood, environmental adversity, food security and poverty); (5) social (mother's social norms and social capital, home food environment). ABM-generated synthetic data will be compared to existing longitudinal empirical data from the MAVAN cohort (lead by Co-I Robert Levitan and Michael Meaney), a sample of mothers and their children observed in a longitudinal within-subject design from the time of pregnancy and birth to examine gene-environment interactions and neurodevelopment. The specific aims of the research proposed here, over the course of the 5- year project, are: (1) to construct separate agent-based "modular" models for each of five levels of analysis relevant to childhood obesity-genetic, neuro-cognitive, psychological, family, and social/environmental--using early childhood data (0-4 years), (2) to integrate multiple modules of the ABM and explore feedback loops between levels, (3) to test the predictive ability of the modular ABM for the key transitional period of 5-7 years, and (4) to provide further validation of the modular ABM models using functional data analysis. The complete model will enable subsequent exploration of potential policy design of interventions, as well as projection of the implications of identified trends. The modular ABM methodology developed in this project would also be of use for the study of other, similarly complex problems. In brief, then, this project will both develop a novel multilevel methodology (modular agent-based computational modeling) and apply it to the study of childhood obesity to improve our understanding of the multilevel determinants of childhood obesity and help design more effective multilevel interventions that consider the range of biological, family, community, socio-cultural, environmental, policy, and macro-level economic factors that influence diet and physical activity in children. PUBLIC HEALTH RELEVANCE: This project will develop a novel multi-level methodology (modular agent-based computational modeling; ABM) and apply it to the study of childhood obesity to improve our understanding of the multilevel determinants of childhood obesity, with a focus on eating. ABM-generated synthetic data will be compared to existing longitudinal empirical data from a cohort of low socio-economic status mothers and their children observed in a longitudinal within-subject design from the time of pregnancy and birth to examine gene-environment interactions and neurodevelopment. The outcome will help design more effective multilevel interventions that consider the range of biological, family, community, socio-cultural, environmental, policy, and macro-level economic factors that influence diet and body weight in children.