README.md

In mbsohn/cmmb: Compositional Mediation Model for Binary Outcomes

README

Michael B. Sohn 7/05/2021

CMMB: Compositional Mediation Model for Binary Outcomes

The cmmb function estimates direct and indirect effects of treatment on binary outcomes mediated through a compositional mediator that consists of multiple components. For detailed information about the arguments, please see the documentation for cmmb().

Note: the number of components can be high-dimensional. However, it will be computationally intensive to run it with high-dimensional components as it tests the indirect effect using debiased bootstrap estimates based on 2,000 (in default) random samplings.

Installation

install.packages(“devtools”)

devtools::install_github(“mbsohn/cmmb”)

Example: estimate direct and total indirect effects

### load functions used in data generation and performance comparison
library(cmmb)
source("comparators.R")
### Generate a simulated dataset of 50 samples and 5 taxa
set.seed(2021)
### The gen.cmm.sim.data function simulates data based on the logistic normal distribution. 
### The first argument takes the number of samples and the second takes the number of components
### (taxa). By default, both direct and indirect effects are significantly different from zero.
sim.dat <- gen.cmm.sim.data(50, 5)
### Run CMM for binary outcomes
rslt <- cmmb(Y=sim.dat$Y, M=sim.dat$M, tr=sim.dat$tr, X=sim.dat$X)
rslt

## $total
##       Estimate Lower Limit Upper Limit
## DE   0.1831665   0.1206513   0.2885970
## TIDE 0.2585639   0.1525723   0.3321615
## 
## $cwprod
##           Estimate Lower Limit Upper Limit
## taxon1  0.45250088   0.0643407   0.7435297
## taxon2  0.00179244  -0.2030688   0.1593904
## taxon3 -0.20841877  -0.5068055   0.1638699
## taxon4  0.61865737  -0.1058526   1.1229078
## taxon5  0.51405582  -0.4161137   1.1703481
## 
## attr(,"class")
## [1] "cmmb"

### Plot products of component-wise path coefficients
plot_cw_ide(rslt)

Example: sensitivity analysis for the total indirect effect

rslt.sa <- cmmb(Y=sim.dat$Y, M=sim.dat$M, tr=sim.dat$tr, X=sim.dat$X, ForSA=TRUE)
### Plot sensitivity of the estimated total indirect effect
plot_cmmb_sa(rslt.sa)

Performance comparison

Note: By default, the gen.cmm.sim.data function simulates data such that both direct and indirect effects are significantly different from zero.

sim.dat <- gen.cmm.sim.data(50, 5)
PCS(Y=sim.dat$Y, M=sim.dat$M, tr=sim.dat$tr, X=sim.dat$X)

##       Estimate Lower Limit Upper Limit
## DE   0.8585578   -1.011952    1.807630
## TIDE 0.5737664   -2.550865    2.263679

PCP(Y=sim.dat$Y, M=sim.dat$M, tr=sim.dat$tr, X=sim.dat$X)

##        Estimate Lower Limit Upper Limit
## DE   0.16603048  0.09844733  0.34177689
## TIDE 0.01797279 -0.15262002  0.07936442

cmmb(Y=sim.dat$Y, M=sim.dat$M, tr=sim.dat$tr, X=sim.dat$X)$total

##       Estimate Lower Limit Upper Limit
## DE   0.1615200  0.11897322   0.2372022
## TIDE 0.1003837  0.02902102   0.1425433

mbsohn/cmmb documentation built on Dec. 21, 2021, 3:56 p.m.