Using Social Network Analysis to Delineate Mechanisms of Behaviour Change in Peer-led Interventions.

基于社会网络的干预措施利用有影响力的个人来改变健康行为，并被认为是基于个人的战略的有希望的替代方案。然而，这些干预措施的设计取决于识别有影响力的同行(例如，“同行领袖”)的能力。各种网络干预策略的大规模全面试验通常在成本和时间上都是令人望而却步的。因此，该项目将采用基于模拟的计算模型来克服这一障碍。该项目将结合社会网络分析和基于代理的建模，比较在基于网络的干预措施中选择有影响力的个人的一系列策略的有效性(例如，根据友谊的受欢迎程度、人脉、有效领导者的同行提名)，并评估每种策略的有效性如何根据模拟的环境背景而不同。面向行为者的随机模型(SAOM)将适用于两个针对13-16岁年轻人的大型社会网络干预的数据，即A停止在学校吸烟试验(n=7,730)(ASSISH)和性传播感染和性健康试验(n=1,376)(STASH)，以提供网络和行为变化的经验性参数。基于代理的模型将模拟不同的干预场景。同行领导者的选择将首先受到操纵，以评估哪种策略导致的行为变化最大。然后，对社会(例如，网络结构、关系强度)和环境背景(例如，学生的人口统计)的操纵进行比较，以评估干预策略和更广泛的背景之间的相互作用。社会网络模型和这些模型的基于代理的扩展非常适合于评估网络干预，因为它们能够以经验数据为基础，并模拟观察数据的反事实。研究结果将告知如何根据个别学校的特定背景特征定制社交网络干预措施，并支持未来学校健康干预措施的发展。

Social network-based interventions capitalize on influential individuals to alter health behaviour, and are recognised as promising alternatives to individual-based strategies. However, the design of these interventions rests on the ability to identify influential peers (e.g., 'peer leaders'). Large, full-scale trials of varying network intervention strategies are often prohibitive in cost and time. Thus, this project will employ simulation-based computational models to overcome this barrier. Using a combination of social network analysis and agent-based modelling, the project will compare the effectiveness of a range of strategies for selecting influential individuals in network-based interventions (e.g., selection based on friendship popularity, connectedness, peer nomination of effective leaders), and assess how the effectiveness of each strategy differs according to simulated environmental context. Stochastic actor-oriented models (SAOMs) will be fit to data from two large social network interventions with young people, ages 13-16, the A Stop Smoking in Schools Trial (n=7,730) (ASSIST), and the Sexually Transmitted Infections and Sexual Health trial (n = 1,376) (STASH), to provide empirically-grounded parameters of network and behaviour change. Agent-based models will simulate varying intervention scenarios. Peer leader selection will be manipulated first, to assess which strategy results in the largest behaviour change. Manipulations to the social (e.g., network structure, relationship strength) and environmental context (e.g., demographics of students) will then be compared to assess the interplay between intervention strategy and wider context. Social network models, and agent-based extensions to these models, are well-suited to evaluate network interventions given their ability to be grounded in empirical data, and simulate counterfactuals to the observed data. The findings will inform how social network interventions can be tailored according to the specific contextual features of individual schools, and support the development of future schools health interventions.