In LindaValeri/CMAverse: Causal Mediation Analysis

We illustrate the general workflow of the CMAverse package by a quick example. The general workflow is:

1. Plot the DAG of causal relationships using cmdag.

2. Estimate causal effects and make inferences using cmest.

3. Conduct sensitivity analysis for unmeasured confounding and measurement error using cmsens.

Firstly, let's load the package.

library(CMAverse)

Next, we simulate some data and plot the DAG. The simulated dataset contains a binary exposure, a binary mediator, a continuous mediator, a continuous outcome and two baseline confounders.

set.seed(1)
n <- 100
C1 <- rnorm(n, mean = 1, sd = 1)
C2 <- rbinom(n, 1, 0.6)
pa <- exp(0.2 - 0.5*C1 + 0.1*C2)/(1 + exp(0.2 - 0.5*C1 + 0.1*C2))
A <- rbinom(n, 1, pa)
pm <- exp(1 + 0.5*A - 1.5*C1 + 0.5*C2)/ (1 + exp(1 + 0.5*A - 1.5*C1 + 0.5*C2))
M1 <- rbinom(n, 1, pm)
M2 <- rnorm(n, 2 + 0.8*A - M1 + 0.5*C1 + 2*C2, 1)
Y <- rnorm(n, mean = 0.5 + 0.4*A + 0.5*M1 + 0.6*M2 + 0.3*A*M1 + 0.2*A*M2 - 0.3*C1 + 2*C2, sd = 1)
data <- data.frame(A, M1, M2, Y, C1, C2)

The DAG can be plotted using cmdag.

cmdag(outcome = "Y", exposure = "A", mediator = c("M1", "M2"), 
      basec = c("C1", "C2"), postc = NULL, node = FALSE, text_col = "black")

Then, we estimate causal effects using cmest. We use the regression-based approach for illustration. The reference values for the exposure are set to be 0 and 1. The reference values for the two mediators are set to be 1.

est <- cmest(data = data, model = "rb", outcome = "Y", exposure = "A",
                mediator = c("M1", "M2"), basec = c("C1", "C2"), EMint = TRUE,
                mreg = list("logistic", "linear"), yreg = "linear",
                astar = 0, a = 1, mval = list(1, 1),
                estimation = "imputation", inference = "bootstrap", nboot = 20)

Summarize and plot the results:

summary(est)

ggcmest(est) +
  ggplot2::theme(axis.text.x = ggplot2::element_text(angle = 30, vjust = 0.8))

Lastly, we conduct sensitivity analysis for the results. Sensitivity analysis for unmeasured confounding:

cmsens(object = est, sens = "uc")

Assume that $C_1$ was measured with error. Sensitivity analysis for measurement error using regression calibration with a set of assumed error standard deviations 0.1, 0.2 and 0.3:

me1 <- cmsens(object = est, sens = "me", MEmethod = "rc", 
              MEvariable = "C1", MEvartype = "con", MEerror = c(0.1, 0.2, 0.3))

Summarize and plot the results:

summary(me1)

ggcmsens(me1) +
  ggplot2::theme(axis.text.x = ggplot2::element_text(angle = 30, vjust = 0.8))

Then, assume that the exposure was measured with error. Sensitivity analysis for measurement error using SIMEX with two assumed misclassification matrices:

me2 <- cmsens(object = est, sens = "me", MEmethod = "simex", MEvariable = "A", 
              MEvartype = "cat", B = 5,
              MEerror = list(matrix(c(0.95, 0.05, 0.05, 0.95), nrow = 2), 
                             matrix(c(0.9, 0.1, 0.1, 0.9), nrow = 2)))

Summarize and plot the results:

summary(me2)

ggcmsens(me2) +
  ggplot2::theme(axis.text.x = ggplot2::element_text(angle = 30, vjust = 0.8))

LindaValeri/CMAverse documentation built on July 16, 2024, 11:58 p.m.