Abstract:Mediation analysis in high-dimensional settings often involves identifying potential mediators among a large number of measured variables. For this purpose, a two step familywise error rate (FWER) procedure called ScreenMin has been recently proposed (Djordjilović et al. 2019). In ScreenMin, variables are first screened and only those that pass the screening are tested. The proposed threshold for selection has been shown to guarantee asymptotic FWER. In this work, we investigate the impact of the selection threshold on the finite sample FWER. We derive power maximizing selection threshold and show that it is well approximated by an adaptive threshold of Wang et al. (2016). We study the performance of the proposed procedures in a simulation study, and apply them to a case-control study examining the effect of fish intake on the risk of colorectal adenoma.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively identify potential mediator variables while controlling the familywise error rate (FWER) when conducting mediation analysis in a high - dimensional data environment. Specifically, the researchers proposed a two - stage testing method - ScreenMin, which aims to screen out possible mediator variables from a large number of measured variables and then test these variables after screening. The paper focuses on the impact of the selection threshold on FWER in finite samples and proposes an optimized selection threshold to maximize the test power while ensuring FWER. ### Main problems and methods 1. **Problem background**: - Mediation analysis in high - dimensional data settings often involves identifying potential mediator variables from a large number of measured variables. - Traditional multiple hypothesis testing methods can be very conservative when dealing with such problems, resulting in low test power. 2. **Proposed solutions**: - **ScreenMin method**: First, screen variables by the minimum p - value (\( p_i=\min\{p_{i1}, p_{i2}\} \)), and then test the screened variables. - **Optimization of selection threshold**: The researchers derived the upper bound of FWER in finite samples and proposed an optimized selection threshold that maximizes the test power while ensuring FWER. 3. **Theoretical contributions**: - It is proved that the ScreenMin method can asymptotically control FWER under the independent p - value assumption. - An optimized selection threshold is proposed, which is the minimum threshold that satisfies the FWER constraint. - The effectiveness of the optimized threshold is verified by numerical simulation and compared with the standard Bonferroni method and other methods. 4. **Practical applications**: - The ScreenMin method is applied to a naval colorectal adenoma case - control study to explore the impact of fish intake on potential mediator variables (such as circulating metabolites) of colorectal adenoma risk. ### Key formulas - **Selection threshold**: \[ c\in(0, 1) \] where \( c \) is the selection threshold. - **Adjusted p - value**: \[ p^*_i = \begin{cases} \min\{|S|p_i, 1\} & \text{if } i\in S \\ 1 & \text{otherwise} \end{cases} \] where \( |S| \) is the size of the selection set. - **Upper bound of FWER in finite samples**: \[ \Pr(V\geq1)\leq E\left[\left(1 - \left(1 - P_0\left(\frac{\alpha}{|S|}, c\right)\right)^{|S|}\right)I[|S| > 0]\right] \] - **Constrained optimization problem of optimized selection threshold**: \[ \max_{0 < c\leq\alpha} E\left[\Pr\left(p_i\leq\frac{\alpha}{|S(c)|}, p_i\leq c\right)I[|S(c)| > 0]\right] \] subject to: \[ \Pr(V(c)\geq1)\leq\alpha \] ### Conclusion Through theoretical derivation and numerical simulation, the paper shows the effectiveness and superiority of the ScreenMin method in high - dimensional mediation analysis. The optimized selection threshold can significantly improve the test power while ensuring FWER. In practical applications, this method also shows good performance in the naval colorectal adenoma study.

Optimal two-stage testing of multiple mediators

On optimal two-stage testing of multiple mediators

Optimal exact tests for multiple binary endpoints

Controlling false discovery rate for mediator selection in high-dimensional data

Two-stage Hypothesis Tests for Variable Interactions with FDR Control

Exploratory Mediation Analysis with Many Potential Mediators

Testing High-Dimensional Mediation Effect with Arbitrary Exposure-Mediator Coefficients

Optimal tests of the composite null hypothesis arising in mediation analysis

Post-selection inference for high-dimensional mediation analysis with survival outcomes

A two-stage Fisher exact test for multi-arm studies with binary outcome variables

On limiting behaviors of stepwise multiple testing procedures

Mediation pathway selection with unmeasured mediator-outcome confounding

Powerful Large-scale Inference in High Dimensional Mediation Analysis

Mediation analysis with case–control sampling: Identification and estimation in the presence of a binary mediator

Mediation effects that emulate a target randomised trial: Simulation-based evaluation of ill-defined interventions on multiple mediators

Indirect Effects in Sequential Mediation Models: Evaluating Methods for Hypothesis Testing and Confidence Interval Formation

Speeding up interval estimation for R2-based mediation effect of high-dimensional mediators via cross-fitting

Analysis of error control in large scale two-stage multiple hypothesis testing

Testing Mediation Effects in High-Dimensional Microbiome Data with False Discovery Rate Control

Group lasso based selection for high-dimensional mediation analysis

Improved Tests for Mediation