Methods for large-scale single mediator hypothesis testing: Possible choices and comparisons.

Area of Science:

• Biostatistics
• Statistical Genetics
• Epidemiology

Background:

• Mediation hypothesis testing with numerous mediators presents challenges due to the composite nature of the null hypothesis (αβ = 0).
• Existing methods often simplify or ignore the distinct components of the null hypothesis: α = 0, β ≠ 0; α ≠ 0, β = 0; or α = β = 0.

Approach:

• Reviewed three classes of large-scale, one-at-a-time mediation testing methods for continuous outcomes and mediators.
• Developed the Sobel-comp method, a novel approach within the mixture reference distribution class.
• Conducted extensive simulations to compare six methods based on false positive rates (FPRs) and true positive rates (TPRs).

Key Points:

• Methods employing a mixture reference distribution demonstrated superior control of FPRs at nominal levels.
• These mixture distribution methods also exhibited the highest TPRs under the alternative hypothesis.
• The study applied these methods to analyze DNA methylation sites in the socioeconomic status to glycated hemoglobin pathway using MESA data.

Conclusions:

• The second class of methods, utilizing a mixture reference distribution, is recommended for large-scale mediation hypothesis testing.
• Guidelines for selecting optimal methods are provided.
• An R package, medScan, is available for implementing the evaluated methods.