Skip to contents

qgcomp.multi

qgcomp.multi extends quantile g-computation to settings with two exposure mixtures and their interaction. It is designed for analyses where exposures come in distinct groups and the effect of shifting one mixture may depend on the level of another.

The package fits an outcome regression model, computes predicted counterfactual means under joint interventions on the two mixtures, and then fits a marginal structural model (MSM) to summarize the resulting intervention-response surface.

What the package does

qgcomp.multi currently supports:

  • two-mixture estimation through qgcomp.glm.multi()
  • optional interaction between the two mixture intervention variables
  • quantized analyses with q >= 2
  • original-scale analyses with q = NULL
  • additive and ratio-scale MSM summaries through estimand_scale
  • Wald, percentile, and basic bootstrap intervals for single-fit coefficient summaries
  • MSM-based prediction, plotting, diagnostics, adequacy checks, and confidence regions
  • sensitivity helpers for Monte Carlo size, bootstrap size, and quantization
  • native multiple-imputation pooling through qgcomp.glm.multi.mi()
  • optional bootstrap-level parallel execution
  • broom-style tidy() and glance() methods for single-fit and pooled MI objects

Installation

You can install the development version from GitHub:

devtools::install_github("cmowrer13/qgcomp.multi")

Main quantized workflow

For a standard quantized analysis, fit a two-mixture model with qgcomp.glm.multi():

library(qgcomp.multi)

fit <- qgcomp.glm.multi(
  f = Y ~ X1 + X2 + X3 + W1 + W2 + W3 + C,
  data = dataset,
  mix1 = c("X1", "X2", "X3"),
  mix2 = c("W1", "W2", "W3"),
  interaction = TRUE,
  q = 4,
  B = 100,
  seed = 13
)

summary(fit)
coef(fit)
confint(fit)

In this setting, the MSM coefficients summarize the expected outcome change associated with simultaneous one-quantile shifts in the component exposures of each mixture. For binary and count outcomes, check the reported estimand scale before interpreting the coefficient table.

Original-scale fitting with q = NULL

If quantization is not scientifically appropriate for the analysis, set q = NULL to fit the outcome model on the original analysis scale and build the MSM over a pooled 3 x 3 intervention grid within each mixture.

fit_cont <- qgcomp.glm.multi(
  f = Y ~ X1 + X2 + X3 + W1 + W2 + W3 + C,
  data = dataset,
  mix1 = c("X1", "X2", "X3"),
  mix2 = c("W1", "W2", "W3"),
  interaction = TRUE,
  q = NULL,
  centering = "median",
  B = 100,
  seed = 13
)

summary(fit_cont)
coef(fit_cont)
confint(fit_cont)

With q = NULL, the MSM coefficients are not one-quantile effects. They are defined with respect to the original-scale intervention coding used to fit the MSM, so interpretation should stay tied to the actual intervention grid.

Model interpretation

The main estimands are the coefficients of the fitted marginal structural model:

  • psi1: Mixture 1 main effect
  • psi2: Mixture 2 main effect
  • psi1:psi2: interaction between the two mixtures when interaction = TRUE

These are summaries of the intervention-response surface implied by the fitted outcome model. For causal interpretation, the usual g-computation conditions still matter: consistency, exchangeability, positivity, and adequate outcome model specification.

Beyond the basic fit

The package also includes tools for:

Those features are documented in more detail in the workflow article, the effect-scales article, the function reference, and the fuller README on GitHub.

Where to go next

References

Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. A Quantile-Based g-Computation Approach to Addressing the Effects of Exposure Mixtures. Environmental Health Perspectives. 2020;128(4):047004.