Compare C-Statistics (Concordance) Between Survival Models

Compare C-statistics (concordance statistics) between two survival models, using either bootstrap resampling (Harrell's C) or Uno's C with perturbation-resampling (from the survC1 package). Returns confidence intervals and a p-value for the difference in C-statistics. Useful for evaluating and comparing predictive performance of survival models. Methods implemented for Uno's C are described in Uno et al. (2011) .

compareCstat

compareCstat is an R package that compares the C-statistics (concordance statistics) between two fitted survival models (e.g., Cox models) using bootstrap resampling. It returns bootstrapped confidence intervals and a p-value for the difference in C-statistics — useful for model performance comparison.

Installation

You can install the development version of compareCstat from GitHub with:

# install.packages("devtools")
devtools::install_github("Lemonade0924/compareCstat")

Example

This is a basic example which shows you how to solve a common problem:

library(compareCstat)
library(survival)

lung <- survival::lung
# Make sure 0 = censored, 1 = event for Uno C

lung$status <- ifelse(lung$status == 2, 1, 0)
model1 <- coxph(Surv(time, status) ~ age, data = lung)
model2 <- coxph(Surv(time, status) ~ age + sex, data = lung)

compare_c_stat(model1, model2, data = lung, R = 100, method = "Uno", tau = 1 * 365.25)
#>                Model C_Statistic CI_Lower CI_Upper P_Value
#> 1      Model 1 (Raw)      0.5486   0.5010   0.5962    <NA>
#> 2 Model 2 (Extended)      0.5991   0.5519   0.6463    <NA>
#> 3         Difference      0.0505   0.0099   0.0910   0.015