生存资料的中介分析方法评价

发布时间：2018-03-07 01:32

本文选题：生存资料　切入点：中介分析　出处：《南京医科大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着癌症确诊率的提高,越来越多的人关注癌症的危险因素及其预后的影响因素。虽然研究者们能找到不同癌症预后的影响因素,但是这些影响因素如何作用于预后尚不清楚。中介分析作为一种能够探索暴露与结局间因果作用机制的方法,广受关注。目前,可用于癌症预后研究的中介分析方法共有五种,分别是VanderWeele法、Baron-Kenny法、Imai法、Sobel法和Inverse Weight法(IW法)。本研究的目的就是对这五种方法进行评价,为实际应用时方法的选择提供依据。本研究基于模拟试验,产生不同参数设置下的模拟数据进行模拟,分别比较这五种方法的第一类错误、检验效能、准确度、精确度和分析效率。模拟试验结果显示,当暴露(X)与中介因素(M)不存在交互作用的情况下,IW法较其它方法检验效能、估计准确度和精度均低于其他几种方法,Baron-Kenny法和Imai法的分析效率远低于VanderWeele法和Sobel法。Baron-Kenny法和Sobel法虽然模型简单、原理便于理解,但这两种方法存在一定的局限性——不能处理暴露与中介因素存在交互作用以及暴露每变化一个水平对结局的效应大小不相同的情况,而VanderWeele法则能较好地处理这些问题,故应用范围更广。IW法对间接效应的点估计是通过总效应减去直接效应间接得到的,同样不适用于"暴露与中介因素存在交互作用"的情况。故当"暴露与中介因素存在交互作用"的情况下,优先考虑VanderWeele法和Imai法。当暴露(X)与中介因素(M)存在交互作用的情况下,VanderWeele法和Imai法的检验效能均随着样本量、总效应、中介比的增大而增大,随着删失比的增大而减小;随着暴露(X)与中介因素(M)之间相关系数的增大而先增大后减小。VanderWeele法的检验效能受中介交互作用比(X:M,%)的影响较小,而Imai法则随着(X:M,%)增大而逐渐减小。无论在何种参数情况下,VanderWeele法的检验效能都不低于Imai法。且VanderWeele法的估计准确度受样本量、总效应、相关系数、中介交互作用CX:M,%)、中介比(M%)和删失比(cen%)的影响很小,其估计精度也远高于IW法。加上VanderWeele法可以得到间接效应、直接效应和总效应的估计值,分析效率也高,故当X和M存在交互作用的时候,VanderWeele法是首先推荐的一个方法。结论:对于生存资料,不论暴露(X)与中介因素(M)之间是否存在交互作用,不管从第一类错误的控制、检验效能、精确度、准确度或者检验效率考虑,VanderWeele法相较于IW法、Imai法、Baron-Kenny法和Sobel法都更为突出,是一个值得推荐的方法。
[Abstract]:With the rate of cancer diagnosis increasing, more and more people are concerned about the risk factors of cancer and its prognostic factors, although researchers can find the factors that affect the prognosis of different cancers. However, it is not clear how these factors affect the prognosis. As a method to explore the causal mechanism between exposure and outcome, mediation analysis has attracted much attention. At present, there are five kinds of intermediary analysis methods that can be used to study the prognosis of cancer. VanderWeele method is Baron-Kenny method and Inverse Weight method is IW method. The purpose of this study is to evaluate these five methods and to provide a basis for the selection of methods in practical application. Simulation data are generated under different parameter settings, and the first type errors of the five methods are compared, respectively. Test effectiveness, accuracy, accuracy and efficiency of analysis are compared. When there is no interaction between exposure X) and mediating factor M), the effectiveness of IW method is higher than that of other methods. The analytical efficiency of Baron-Kenny method and Imai method is much lower than that of VanderWeele method, Sobel method. Baron-Kenny method and Sobel method although the model is simple and the principle is easy to understand. But these two methods have some limitations-they can't deal with the interaction between exposure and mediating factors, and the difference in the effect of each level of exposure on the outcome, and the VanderWeele rule can better deal with these problems. So the point estimation of indirect effect is obtained by subtracting direct effect from total effect. The same does not apply in the case of "interaction between exposure and mediating factors". Therefore, in the case of "interaction between exposure and intermediary factors", Priority is given to VanderWeele method and Imai method. When there is interaction between X) and the intermediate factor M), the efficiency of van der Weele method and Imai method both increase with the increase of sample size, total effect and intermediary ratio, and decrease with the increase of censored ratio. With the increase of the correlation coefficient between the exposure X) and the intermediate factor M), the test effectiveness of the Vander Weele method increased first and then decreased. The test effectiveness of the Vander Weele method was less affected by the medium interaction than by X: M). However, the Imai rule decreases with the increase of X: M). The efficiency of Vander Weele method is not lower than that of Imai method under any parameters. The accuracy of VanderWeele method is estimated by sample size, total effect and correlation coefficient. The influence of the intermediary interaction, CX: M) and the cenb, is very small, and its estimation accuracy is much higher than that of the IW method. In addition, the VanderWeele method can obtain the estimated value of indirect effect, direct effect and total effect, and the efficiency of analysis is also high. Therefore, when there is interaction between X and M, Vander Weele method is the first recommended method. Conclusion: for survival data, whether or not there is interaction between X) and mediating factor M), whether or not there is interaction between X and M, no matter from the control of the first error, the effectiveness is tested. The Vander Weele method is more prominent than the IW method in comparison with the Baron-Kenny method and the Sobel method. It is a recommended method.
【学位授予单位】：南京医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R73-31

【参考文献】