Table 4.

Methods Summary.

Name and Author	Estimation of global indirect effect	Estimation of mediation contributions	Mediator identification	Y Data Type	Summary
Group 1 Methods
HIMA; Zhang, 2016	Point estimation	Point estimation	Yes	Continuous or binary	Fits the outcome model with the minimax concave penalty. Requires subsequent fitting of ordinary least squares regression to test the statistical significance the mediation contributions.
HDMA; Gao, 2019	Point estimation	Point, interval estimation	Yes	Continuous or binary	Fits the outcome model with the de-sparsified LASSO penalty.
MedFix; Zhang, 2021	Point estimation	Point, interval estimation	Yes	Continuous	Fits the outcome model with the adaptive LASSO penalty. Can also be applied when the exposure is high-dimensional in addition to the mediators.
Pathway LASSO Zhao and Luo, 2022	Point estimation	Point estimation	Yes	Continuous	Fits the outcome model and mediator models with a jointly penalized likelihood, directly applying shrinkage to the mediation contributions $({\alpha }_{\mathit{a}}{)}_j({\beta }_{\mathit{m}}{)}_j$.
BSLMM; Song, 2020	Bayesian point, interval estimation	Bayesian interval estimation	Yes	Continuous	Bayesian mixed-model in which the mediator-outcome associations $({\beta }_{\mathit{m}}{)}_j$ and the exposure-mediator associations $({\alpha }_{\mathit{a}}{)}_j$ are assumed to independently follow sparse normal distributions.
GMM; Song, 2021	Bayesian point, interval estimation	Bayesian interval estimation	Yes	Continuous	Bayesian mixed-model in which the mediator-outcome associations $({\beta }_{\mathit{m}}{)}_j$ and the exposure-mediator associations $({\alpha }_{\mathit{a}}{)}_j$ are assumed to jointly follow a sparse multivariate normal distribution.
Group 3 Methods
PCMA; Huang and Pan, 2016	Point, interval estimation	No	No	Continuous or binary	Applies principal component analysis on the mediator model residuals, transforming the mediators so they are independent. Can be applied when there is A-M interaction in the outcome model.
SPCMA; Zhao, 2019	Point, interval estimation	No	Identifies whether subsets of the mediators are jointly active	Continuous	Similar to PCMA but applies sparse PCA, resulting in transformed mediators that are more interpretable.
HILMA; Zhou, 2020	Point, interval estimation	No	No	Continuous	Uses a debiased penalized regression approach to directly estimate the global indirect effect ${\alpha }_{\mathit{a}}^T{\beta }_{\mathit{m}}$. Can be applied for multiple exposures simultaneously.
Group 3 Methods
HDMM; Chen, 2018	No	No	Nonspecifically identifies groups of active mediators	Continuous	Estimates “directions of mediation” by which the observed mediators can be linearly combined to form latent mediators. The latent mediators replace the true mediators in the analysis.
LVMA; Derkach, 2019	No	No	Identifies inputted mediators associated with latent mediators	Continuous or binary	Reformulates the causal structure of the mediation problem. Assumes that $\mathit{M}$ itself is not responsible for mediation, but rather that the effect of A on Y is mediated by latent, unmeasured factors, $\mathit{F}$, which also cause changes in $\mathit{M}$.