High-Dimensional Mediation Analysis With Confounders in Survival Models

Abstract

Mediation analysis is a common statistical method for investigating the mechanism of environmental exposures on health outcomes. Previous studies have extended mediation models with a single mediator to high-dimensional mediators selection. It is often assumed that there are no confounders that influence the relations among the exposure, mediator, and outcome. This is not realistic for the observational studies. To accommodate the potential confounders, we propose a concise and efficient high-dimensional mediation analysis procedure using the propensity score for adjustment. Results from simulation studies demonstrate the proposed procedure has good performance in mediator selection and effect estimation compared with methods that ignore all confounders. Of note, as the sample size increases, the performance of variable selection and mediation effect estimation is as well as the results shown in the method which include all confounders as covariates in the mediation model. By applying this procedure to a TCGA lung cancer data set, we find that lung cancer patients who had serious smoking history have increased the risk of death via the methylation markers cg21926276 and cg20707991 with significant hazard ratios of 1.2093 (95% CI: 1.2019–1.2167) and 1.1388 (95% CI: 1.1339–1.1438), respectively.

Instrumental variable-based high-dimensional mediation analysis with unmeasured confounders for survival data in the observational epigenetic study

“…To ensure the identification of the mediating effects, there are several assumptions that need to be hold for the methodology proposed in this study ( VanderWeele, 2011 ; Huang and Yang, 2017 ; Yu et al, 2021 ; Tian et al, 2022 ).…”

“…The main R packages used in the current study include "survival," "ncvreg," "ggm," "ivtool," "glmnet," and "boot." The choice of simulation parameters was based on published methodology studies and application studies focusing on the mediating role of epigenetic factors (Luo et al, 2020;Yu et al, 2021;Tian et al, 2022).…”

“…The development of the confounder adjustment methodology has greatly enriched the application of mediating effect analysis ( Coffman, 2011 ; Valeri and VanderWeele, 2013 ). Yu et al (2021) expanded Luo et al’s (2020) approach ( Luo et al, 2020 ) with the PS adjustment to control potential confounders ( Yu et al, 2021 ). Liu et al (2022) proposed a powerful divide-aggregate composite-null test (DACT) for causal mediation effects ( Liu et al, 2022 ).…”

Linear high-dimensional mediation models adjusting for confounders using propensity score method

“…By controlling the propensity score in a proper way like matching, regression, or inverse probability weighting, the confounders could be adjusted, which helps to create a theoretical randomized controlled trial (RCT) ( Rosenbaum and Rubin, 1983 ; D'Agostino, 1998 ) and satisfy the ignorability assumption. Compared with the regression adjustments, propensity score concentrates all covariates into a single “score” variable, which is more flexible and adequate to eliminate confounding bias ( Austin, 2011 ; Yu et al, 2021 ). Previous studies have already applied PS in mediation analysis ( Coffman, 2011 ; Jo et al, 2011 ; Yu et al, 2021 ).…”

“…Compared with the regression adjustments, propensity score concentrates all covariates into a single “score” variable, which is more flexible and adequate to eliminate confounding bias ( Austin, 2011 ; Yu et al, 2021 ). Previous studies have already applied PS in mediation analysis ( Coffman, 2011 ; Jo et al, 2011 ; Yu et al, 2021 ). However, there is still a lack of insights into the appropriate utilization of PS for adjusting confounders in HIMA under continuous (or binary) outcomes.…”