Objective: To introduce and compare four analysis methods of multiple parallel mediation model, including pure regression method, method based on inverse probability weighting, extended natural effect model method and weight-based imputation strategies. Methods: For the multiple parallel mediation model, the simulation experiments of three scenarios were carried out to compare the performance of different methods in estimating direct and indirect effects in different situations. Dataset from UK Biobank was then analyzed by using the four methods. Results: The estimation biases of the regression method and the inverse probability weighting method were relatively small, followed by the extended natural effect model method, and the estimation results of the weight-based imputation strategies were quite different from the other three methods. Conclusions: Different multiple parallel mediation analysis methods have different application situations and their own advantages and disadvantages. The regression method is more suitable for continuous mediator, and the inverse probability weighting method is more suitable for binary mediator. The extended natural effect model method has better performances when the residuals of two parallel mediators are positively correlated and the correlation degree is small. The weight-based imputation strategies might not be appropriate for parallel mediation analysis. Therefore, appropriate methods should be selected according to the specific situation in practice.
目的： 介绍4种多重并行中介模型的分析方法，包括纯回归法、逆概率加权法、扩展的自然效应模型和基于权重的填补法，并对其进行探讨和比较。 方法： 针对多重并行中介模型，通过3种情境的模拟试验比较不同方法在不同情境下估计直接效应和间接效应的表现，并应用英国生物样本库的数据集进行实例分析。 结果： 模拟试验和实例分析结果显示纯回归法和逆概率加权法对各效应的估计偏倚较小，扩展的自然效应模型次之，基于权重的填补法与另外3种的估计结果差异较大。 结论： 不同的多重并行中介分析方法有不同的适用情境以及各自的优缺点，纯回归法更适用于连续中介的情形，逆概率加权法更适用于二分类中介的情形，扩展的自然效应模型在用于两个并行中介的残差呈正相关且相关程度较小时更佳，而基于权重的填补法可能并不适用于并行中介的情形，因而实际应用时应根据具体情境选择合适的方法。.