Exploring Health Risk Factors: A Causal Mediation Analysis

Introduction

In studies of epidemiology and health-related science, the overall effect of a certain risk factor or exposure on health related outcome is of interest. Once this effect has been confirmed, interest in investigating the possible biological mechanism will grow. By decomposing the overall effect into several pathways according to the involvement of different mediators of interest, path analysis and mediation analysis have been widely adapted to investigate the mechanism 1. In recent decades, causal mediation analysis, a branch of mediation analysis, uses the causal theory to explicitly define effects of interest based on counterfactual outcome model (also named potential outcome model) 2-4, expending mediation analysis to generalized

Several strategies are available for dealing with this problem. One approach is to pool all ordered mediators as a single mediator of interest and decompose the total effect into effect through or not through this pooled mediator 17, 18. Another approach is to assume a strong cross-world assumption to measure the upper and lower bounds of PSE through sensitivity analysis, while the point estimation of PSE is still unable to be obtained 19. Lin and VanderWeele have proposed an interventional approach to estimate analogues of path-specific effects under no unmeasured confounding assumptions with a regression based approach and a corresponding SAS code 20. However, several limitations of this method are noted. One is that the outcome has to follow an ordinary linear regression model, and that it cannot be adapted to non-linear or generalized linear models. Moreover, unlike the analysis of overall effect, the analytical solutions for all PSEs estimation vary substantially in different models even the outcome follows linear model. Therefore, the software based on has only few model