README.md

In SharonLutz/reverseMA: Runs mediation analysis with the mediator and outcome both correctly and incorrectly specified.

reverseMA

Examines the indirect and direct path in mediation analysis when the mediator and outcome are either correctly or incorrectly specified.

Installation

install.packages("mediation") # mediation package must be installed first 
install.packages("devtools") # devtools must be installed first

devtools::install_github("SharonLutz/reverseMA")

Input

For n subjects (input n) for a given minor allele frequency (input pX), the SNP X is generated from a binomial distribution.

Then, the mediator M is generated from a normal distribution with the variance (input varM) and the mean as follows:

E[M] = γo + γX X

where gamma0 and gammaX are inputted by the user.

Next, the outcome Y is generated from a normal distribution with the variance (input varY) and the mean as follows:

E[Y] = βo + βX X + βM M

where beta0, betaX, and betaM are inputted by the user. beta0 and betaX are only allowed to be a single value (i.e. beta0=0), but betaM should be inputted as a vector (i.e. betaM = c(0.1, 0.2, 0.3)).

Output

After the SNP X, mediator M, and outcome Y are generated, then the reverseMA function runs mediation analysis twice: (1) the mediator and outcome are correctly specified, (2) the mediator and outcome are switched. A plot is then produced where the x axis is the values of betaM (i.e. the relationship between the mediator and outcome) and the y-axis is the proportion of simulations where there was a significant direct or indirect effect.

Example 1:

For example 1, we generate a direct effect of the exposure X on the outcome Y (betaX=0.2), but no indirect effect of the exposure X on the outcome Y through the mediator M (gammaX=0). For this example, there are 1000 subjects (n=1000), a MAF of 0.2 (pX=0.2), M is generated from a normal distribution with a variance of 1 (varM=1), Y is generated from a normal distribution with a variance of 1 (varY=1), and the strength of the association between M and Y vairaes from 0.1 to 0.3 (betaM = c(0.1, 0.2, 0.3)). The code below runs this example.

library(reverseMA)

# Direct effect of the exposure X on the outcome Y
# But no indirect effect of the exposure X on the outcome Y through the mediator M

 reverseMA(n = 1000, pX = 0.2, gamma0 = 0, gammaX = 0, varM = 1, beta0 = 0, betaX = 0.2, 
betaM = c(0.1, 0.2, 0.3), varY = 1, nSim = 100, nSimImai = 100, SEED = 1, plot.pdf = T, 
plot.name = "reverseMAplotDirect.pdf", alpha_level = 0.05)

Output 1

As seen in the plot below, there is a significant direct effect if the mediator and outcome are correctly specified (directNR), but counter to how the data was generated, there is a significant indirect effect if the mediator and outcome are incorrectly specified (indirectR).

Example 2:

Example 2 is similar to Example 1 except we now generate an indirect effect of the exposure X on the outcome Y through the mediator M(gammaX=0.2), but no direct effect of the exposure X on the outcome Y (betaX=0). The code below runs this example.

library(reverseMA)

# Indirect effect of the exposure X on the outcome Y through the mediator M
# But no direct effect of the exposure X on the outcome Y

reverseMA(n = 1000, pX = 0.2, gamma0 = 0, gammaX = 0.2, varM = 1, beta0 = 0, betaX = 0, 
betaM = c(0.1, 0.2, 0.3), varY = 1, nSim = 100, nSimImai = 100, SEED = 1, plot.pdf = T, 
plot.name = "reverseMAplotIndirect.pdf", alpha_level = 0.05)

Output 2

As seen in the plot below, there is a significant indirect effect if the mediator and outcome are correctly specified (indirectNR), but counter to how the data was generated, there is a significant direct effect if the mediator and outcome are incorrectly specified (directR).

Example 3:

Example 3 is similar to Example 1 except we generate both a direct (betaX=0.2) and indirect effect (gammaX=0.2) of the exposure X on the outcome Y for a mediator M.

library(reverseMA)

#Both a direct and indirect effect of the exposure X on the outcome Y for a mediator M
 reverseMA(n = 1000, pX = 0.2, gamma0 = 0, gammaX = 0.2, varM = 1, beta0 = 0, betaX = 0.2, 
betaM = c(0.1, 0.2, 0.3), varY = 1, nSim = 100, nSimImai = 100, SEED = 1, plot.pdf = T, 
plot.name = "reverseMAplotBoth.pdf", alpha_level = 0.05)

Output 3

As seen in the plot below, there is both a significant direct and indirect effect of the exposure X on the outcome Y for a mediator M.

Reference

Lutz SM, Sordillo J, Hokanson JE, Wu AC, Lange C. (2020) The Effects of Misspecification of the Mediator and Outcome in Mediation Analysis. Gen Epi. In Press.

Speed Improvements using C++ Bindings and Threading

A version of this package called reverseMAthread implements support for threading and a version of the mediation algorithm implemented using RcppEigen which can provide a significant boost in speed akin to multiprocessing but with less of a memory footprint.

Find it here ReverseMAthread

SharonLutz/reverseMA documentation built on June 15, 2020, 7:20 p.m.