A Regression-Based Approach for Mediation Analysis with Censored Data

Abstract

The regression method is one of the most applied approaches to estimate the direct and indirect effect of the causal variable (X) to an outcome (Y) with any mediating variable (M) in their mechanism. Censored or limited data on any one of the inputs, mediators, and output often observed in a randomized trial with a specific follow-up time or with certain conditions and limitations caused by a particular setting. In this study, several regression strategies were employed to estimate each different path parameter, i.e., the effect of X on M(a), the effect of M on Y (b), and the effect of X on Y (c’) for a single mediator model. A simulation study was conducted under various conditions to assess the performance of the proposed strategy while dealing with a censored mediator. A continuous mediator and outcome are observed in this study so that the product method can be applied to understand the estimation of the indirect effect. Based on the results, it is clearly shown that a higher censoring percentage in the mediating variable yields higher bias estimate. By applying Tobit model in M-regression can minimize the bias of estimate effect with higher censoring percentage. Furthermore, it can conclude that the proposed strategy can effectively perform the estimation of indirect effect with a censored mediator.