| Summary: | 碩士 === 國立交通大學 === 統計學研究所 === 107 === Data truncated by death often happens in a randomized trial with a follow-up study which examines the effect of exposure to outcome. This event might cause incomplete information, undefinable values for any variables involved in the analysis. A traditional mediation analysis can be used to investigate the effect on truncated-by-death case, by merely omitting incom-plete observations, or in other words, limiting the analysis to individuals who survived. How-ever, such an analysis potentially yields a biased estimate. In this study, a regression-based approach was used as an analytic approach for estimating the extended definition of causal effect for “truncated-by-death” case. The method handles binary or continuous mediator with a potential binary outcome. A simulation study was conducted under various conditions to illustrate the performance of the proposed approach. Based on the simulation study, it shows that the complete case method failed to estimate the causal mediation effect. Furthermore, the bias value and coverage rate show that the proposed approach can effectively perform the causal effect of the probability of survivors in the population.
|
|---|
