getCrudeAndAdjustedModelData: This function helps with printing regression models

Description

This function is used for getting the adjusted and unadjusted values for a regression model. It takes a full model and walks through each variable, removes in the regression all variables except one then reruns that variable to get the unadjusted value. This functions not intended for direct use, it's better to use printCrudeAndAdjustedModel() that utilizes this function.

Usage

getCrudeAndAdjustedModelData(
  model,
  level = 0.95,
  remove_interaction_vars = TRUE,
  remove_strata = FALSE,
  remove_cluster = FALSE,
  var_select,
  ...
)
# S3 method for getCrudeAndAdjustedModelData
[(x, i, j, ...)

Value

Returns a matrix with the columns:

c("Crude", "2.5 %", "97.5 %", "Adjusted", "2.5 %", "97.5 %"). The row order is not changed from the original model. The percentages can vary depending on the set level.

Arguments

model: The regression model
level: The confidence interval level
remove_interaction_vars: Removes the interaction terms as they in the raw state are difficult to understand
remove_strata: Strata should most likely not be removed in the crude version. If you want to force the removal of stratas you can specify the remove_strata = TRUE
remove_cluster: Cluster information should most likely also retain just as the remove_strata option. Clusters are sometimes used in cox regression models, cluster()
var_select: A vector with regular expressions for choosing what variables to return (the same format as for the order argument in printCrudeAndAdjustedModel() call). It can be useful when working with large datasets only to report a subsection of all tested variables. This makes the function both run faster and the data presentation more concise.
...: Not used

Details

This function saves a lot of time creating tables since it compiles a fully unadjusted list of all your used covariates.

If the model is an exponential poisson/logit/cox regression model then it automatically reports the exp() values instead of the original values

The function skips by default all spline variables since this becomes very complicated and there is no simple $$\beta$$ to display. For the same reason it skips any interaction variables since it's probably better to display these as a contrast table.

Note that the rms regression has a separate function that uses the rms:::summaryrms function that returns a matrix that is then pruned.

Examples

Run this code

# simulated data to use
set.seed(10)
ds <- data.frame(
  ftime = rexp(200),
  fstatus = sample(0:1, 200, replace = TRUE),
  x1 = runif(200),
  x2 = runif(200),
  x3 = runif(200),
  x4 = runif(200),
  x5 = runif(200)
)

library(rms)
dd <- datadist(ds)
options(datadist = "dd")

s <- Surv(ds$ftime, ds$fstatus == 1)
fit <- cph(s ~ x1 + x2 + x3, data = ds)

data_matrix <- getCrudeAndAdjustedModelData(fit)

print(data_matrix)

# If we have interaction then those variable are not
# reported
fit <- cph(s ~ x1 + x2 + x3 + x4 * x5, data = ds)
data_matrix <- getCrudeAndAdjustedModelData(fit)

print(data_matrix)