compare.AUC.ht: A bootstrap-based hypothesis test to select the best number...

In CatPredi: Optimal Categorisation of Continuous Variables in Prediction Models

A bootstrap-based hypothesis test to select the best number of categories for a continuous predictor variable in a logistic regression model

Description

Compares two objects of class "catpredi" to evaluate the significance of the improvement in model performance (in terms of the AUC) by adding k+1 cut-off points to the predictor variable.

Usage

compare.AUC.ht(
  obj1,
  obj2,
  level = 0.95,
  nb = 100,
  parallel = TRUE,
  plot = TRUE
)

Arguments

obj1	An object inheriting from class "catpredi" for k number of cut points.
obj2	An object inheriting from class "catpredi" for k+1 number of cut points.
level	The confidence level required for the hypothesis test. By default level = 0.95.
nb	Number of bootstrap resamples. By default nb = 100
parallel	A logical value. if TRUE the bootstrap is processed in parallel.
plot	A logical value. if TRUE the density plot for the bootstrap statistic is provided.

Value

This function returns an object of class "compare.AUC.ht" with the following components:

t.null

test statistic, with the difference of the AUCs for the two objects.

t.boot

a vector with the nb bootstrap statistics.

t.null

empirical level-percentile of the bootstrap statistics vector.

Author(s)

Irantzu Barrio, Inmaculada Arostegui, Javier Roca-Pardinas and Xabier Amutxastegi.

References

I Barrio, J Roca-Pardinas and I Arostegui (2021). Selecting the number of categories of the lymph node ratio in cancer research: A bootstrap-based hypothesis test. Statistical Methods in Medical Research, 30(3), 926-940.

See Also

catpredi, comp.cutpoints

Examples

library(CatPredi)
## Not run: 
  set.seed(127)
  #Simulate data
  n = 100
  #Predictor variable
  xh <- rnorm(n, mean = 0, sd = 1)
  xd <- rnorm(n, mean = 1.5, sd = 1)
  x <- c(xh, xd)
  #Response
  y <- c(rep(0,n), rep(1,n))
  # Data frame
  df <- data.frame(y = y, x = x)
  # Select 2 optimal cut points using the AddFor algorithm. Correct the AUC
    res.addfor.k2 <- catpredi(formula = y ~ 1, cat.var = "x", cat.points = 2,
                              data = df, method = "addfor", range=NULL, correct.AUC=TRUE,
                              control=controlcatpredi(grid=20))
  # Select 3 optimal cut points using the AddFor algorithm. Correct the AUC
    res.addfor.k3 <- catpredi(formula = y ~ 1, cat.var = "x", cat.points = 3,
                              data = df, method = "addfor", range=NULL, correct.AUC=TRUE,
                              control=controlcatpredi(grid=20))
    comp <-  comp.cutpoints(res.addfor.k2, res.addfor.k3, V = 10)

  # Select 1 optimal cut points using the BackAddFor algorithm.
    res.backaddfor.k1 <- catpredi(formula = y ~ 1, cat.var = "x", cat.points = 1,
                                  data = df, method = "backaddfor", range=NULL, correct.AUC=FALSE)
  # Select 2 optimal cut points using the BackAddFor algorithm.
    res.backaddfor.k2 <- catpredi(formula = y ~ 1, cat.var = "x", cat.points = 2,
                                  data = df, method = "backaddfor", range=NULL, correct.AUC=FALSE)
  # Test if k=1 cut-off points is enough to categorise x
    comp.k1.k2 <-  compare.AUC.ht(res.backaddfor.k1, res.backaddfor.k2)

## End(Not run)

CatPredi documentation built on May 8, 2026, 9:07 a.m.