基于随访研究资料的中介模型因果推断分析方法研究

Recent studies have shown that the assessment of mediation is an important tool for addressing the criticism of the limitations of ‘black box’ epidemiology and ‘white box’ epidemiology. By using mediation analysis to open the black box and improve white box theory, underlying mechanisms of the observed associations can be described and causal inference further improved. Follow-up study is considered as a prospective Study, and the outcome of interest will often be time-to-event (or, in other words, survival time). As is well known, survival analysis is one of the important research areas of modern medicine statistical analysis. However, since survival times are non-normal and will typically be right censored, the existing causal mediation analysis with survival data cannot meet the actual needs. The main work of this paper includes two aspects: 1) several methods have been proposed to adjust for the bias in mediation analysis with time-to-event outcome. In this part, bias correction methods for the mediation analysis with survival time are improved, and then using Bayesian method to calibration measurement bias is proposed. 2) A novel way of quantifying mediation in a survival context is proposed to estimate the effect mediated through multiple mediators and the effects through other pathways. In this part, we provide a sensitivity analysis technique and a novel analytic strategy for mediation analysis with multiple mediators that are applicable even if there are time-to-event outcome. What is more, new software for the mediation analysis will be developed. The methods and software will be validated and improved by means of real data analysis. This study will promote the development of statistical theory and methods of the causal mediation analysis in epidemiology.

中介分析是一种替代“黑箱理论”和完善“白箱理论”的重要分析方法，该法考虑研究因素之间的联系，进一步改善因果推断结论。随访研究作为一种前瞻性研究，同时考虑随访对象的结局和“生存时间”。生存分析是现代医学统计分析的重要研究领域，由于生存数据的复杂机制和结构，现有的基于生存结局的中介分析方法不能满足实际需求。本课题就基于生存时间数据的中介分析偏倚校正和多中介分析两方面的问题进行系统研究(推导基于生存结局的中介分析偏倚校正公式，提出应用Bayes思想考虑“借用”外部信息校正测量信息偏倚，考虑将常见结局变量转化为二分类变量进行中介分析敏感性评估，根据中介变量间相关关系指导多中介分析策略选择)，并结合计算机模拟和实际资料验证环节对统计分析方法进行系统评价，并开发相应的计算机软件，进而推动流行病学中介效应因果推断的统计理论和方法的发展。

本课题以中介模型的理论与应用研究为背景，研究了以下三个主要内容：.①考虑肿瘤临床试验数据的复杂特征，推导出评价肿瘤临床试验替代终点有效性的中介效应估计公式，通过统计模拟试验，深入评价了这两种中介模型(“Aalen&AFT模型”法和“AFT&AFT模型”法)的统计学性质，结果表明推荐使用“AFT&AFT模型”法。.②为评估中介因子在大气污染物暴露导致不良妊娠结局的潜在调控作用研究积累经验，探讨了宁波市空气污染物与新生儿出生体重之间的关联性。发现PM2.5、SO2、CO、O3浓度暴露增加与低出生体重存在关联。本研究为开展基于中介模型的环境暴露-多组学数据-不良出生结局分析提供了建模经验。.③此外，为积累多组学数据实验研究和建模经验，分别开展circRNA相对表达量的改变与原发性高血压的关联性研究，首次发现circ_0126991相对表达量高表达是原发性高血压的保护因素；开展基于DNA甲基化的人工神经网络模型在原发性高血压诊断中的应用研究，发现相较于传统的Logistic模型，BP-ANN模型在原发性高血压的诊断和预测研究中表现更加出众。.