Network modeling and robust estimation of the intraclass correlation coefficient to inform the design and analysis of cluster randomized trials for infectious diseases

7.项目总结/摘要 迫切需要支持产生高质量证据的研究，以告知临床决策 制作。整群随机试验（CRT）达到了最高标准的证据， 社区一级防治传染病干预战略的有效性。但有必要 开发新的方法来改善CRT的设计和分析，因为独特而复杂的分析方法， 在这种情况下会出现挑战。一个这样的问题涉及组内相关系数（ICC），程度 一个群体中的个体之间比其他群体中的个体之间更相似。 CRT的设计和分析必须考虑ICC。缺乏关于国际刑事法院的准确信息 危害CRT的功率，导致分析方法的次优选择，并使 研究结果的解释。然而，很难获得关于国际刑事法院的可靠资料。一个强大的和高效率 估计ICC的方法是基于二阶广义估计方程。然而，其使用 已经受到相当大的计算负担和与现有算法相关的差的收敛速度的限制。 算法求解这些方程。第一个目标是解决这些计算挑战。缺失数据 无处不在，并可能导致偏见和效率损失。第二个目标是开发新的鲁棒和 在信息缺失数据存在的情况下估计ICC的有效方法。对于传染病， 潜在的接触/传输网络引起复杂的相关结构。第三个目标是 开发网络和流行病模型来预测ICC。将开发方便用户的软件， 新方法的实施。所提出的方法的一个直接应用是它们应用于 博茨瓦纳综合预防项目，以改善对干预效果的估计， 可靠的ICC估计，用于在同一人群中设计未来的CRT。所提出的方法可以应用于 其他正在进行和未来的标准化小组，更广泛地说， 也引起了极大的兴趣。拟议的研究意义重大，因为成功解决这些问题将 提高设计高效和功率良好的CRT的能力，并提高估计 干预策略。创新在于改进计算算法的发展， 深度学习的方法，半参数效率理论的使用，以及网络的整合 建模、流行病建模和统计推断。研究结果将使双方受益。 正在进行的和未来的CRT，允许更有效地利用资源，并最终加快控制 传染病

7. Project Summary/Abstract There is an urgent need to support research that generates high-quality evidence to inform clinical decision making. Cluster randomized trials (CRTs) achieve the highest standard of evidence for the evaluation of community-level effectiveness of intervention strategies against infectious diseases. However, there is a need to develop new methods to improve the design and analysis of CRTs because unique and complicated analytical challenges arise in such settings. One such issue relates to the intraclass correlation coefficient (ICC), the degree to which individuals within a community are more similar to one another than to individuals in other communities. Design and analysis of CRTs must take into account the ICC. Lack of accurate information on the ICC jeopardizes the power of CRTs, leads to suboptimal choices of analysis methods and complicates the interpretation of study results. However, reliable information on the ICC is difficult to obtain. A robust and efficient approach for estimating ICCs is based on the second-order generalizing estimating equations. However, its use has been limited by considerable computational burden and poor convergence rates associated with the existing algorithms solving these equations. The first aim addresses these computational challenges. Missing data are ubiquitous and can lead to bias and loss of efficiency. The second aim proposes to develop novel robust and efficient methods for estimating ICCs in the presence of informative missing data. For infectious diseases, the underlying contact/transmission networks give rise to complicated correlation structure. The third aim is to develop network and epidemic models to project the ICC. User-friendly software will be developed to facilitate the implementation of new methods. An immediate application of the proposed methods is their application to the Botswana Combination Prevention Project to improve the estimation of intervention effect and to generate reliable ICC estimates for designing future CRTs in the same population. The proposed methods can be applied to other ongoing and future CRTs, and more broadly, to longitudinal studies and agreement studies where ICCs are also of great interest. The proposed research is significant, because success in addressing these issues will improve the ability to design efficient and well-powered CRTs and the precision in estimating the effects of intervention strategies. Innovation lies in the development of improved computing algorithms adapting approaches from deep learning, the use of semiparametric efficiency theory, and the integration of network modeling, epidemic modeling and statistical inference. The results of the proposed research will benefit both ongoing and future CRTs, permit more efficient use of the resources, and ultimately expedite the control of infectious diseases.