Quantize exposure variables for two-mixture quantile g-computation

Converts continuous exposure variables belonging to two mixtures into discrete quantile categories. Each exposure is partitioned into q approximately equal-sized groups using empirical quantiles, and replaced with integer values ranging from 0 to q - 1. This transformation places mixture components on a common scale, enabling interpretation of model parameters as effects of simultaneous one-quantile increases in each component.

Usage

quantize_mixtures(data, mix1, mix2, q = 4)

Arguments

data: A data frame containing the exposure variables and any additional variables (e.g., outcome or covariates).
mix1: A character vector giving the names of the variables in the first exposure mixture.
mix2: A character vector giving the names of the variables in the second exposure mixture.
q: Integer greater than or equal to 2 giving the number of quantiles used to discretize each exposure variable, or NULL to skip quantization and return the mixture variables unchanged after validation.

Value

A data frame of the same dimensions as data. When q is numeric, the variables listed in mix1 and mix2 have been replaced by their quantized versions. Quantized variables take integer values in {0, 1, ..., q - 1}. When q = NULL, the returned data are unchanged.

Details

Quantization is a key step in quantile g-computation, as it ensures that exposure variables measured on different scales are made comparable. By transforming each exposure into quantile categories, a one-unit change in a quantized exposure corresponds to a shift of one quantile category. In the fitted two-mixture MSM, this supports the interpretation of the mixture coefficients as summaries of joint one-quantile shifts in all components of a mixture.

Quantiles assign observations to groups of approximately equal size. The resulting categories are then shifted to start at 0 (rather than 1) to align with the intervention levels used in the marginal structural model.

When q = NULL, this function performs no discretization and simply returns data unchanged after checking that the mixture definitions are valid.

Examples

dat <- sim_mixture_data(
  n = 100,
  pA = 3,
  pB = 3,
  rho_within_A = 0.3,
  rho_within_B = 0.3,
  rho_between = 0.2,
  psi1 = 0.5,
  psi2 = 0.3,
  psi12 = 0.2,
  return_quantized = FALSE,
  seed = 123
)

dat_q <- quantize_mixtures(
  data = dat,
  mix1 = c("X1", "X2", "X3"),
  mix2 = c("W1", "W2", "W3"),
  q = 4
)

head(dat_q)
#>            Y X1 X2 X3 W1 W2 W3           C
#> 1  1.7680154  1  2  1  1  0  3  1.07401226
#> 2 -0.5122338  0  1  1  3  0  2 -0.02734697
#> 3  2.8291306  3  3  2  3  3  2 -0.03333034
#> 4  0.2805596  2  3  0  2  1  1 -1.51606762
#> 5  1.1256759  3  0  3  1  1  1  0.79038534
#> 6  3.5903596  1  3  3  3  3  2 -0.21073418

dat_cont <- quantize_mixtures(
  data = dat,
  mix1 = c("X1", "X2", "X3"),
  mix2 = c("W1", "W2", "W3"),
  q = NULL
  )