ROC Curves


A ROC curve (receiver operating characteristic curve) is a graph showing the performance of a classification model at all classification thresholds.

Data

We load the packages we will use :

# libraries
library(pROC)
library(dplyr)
library(ggplot2)
library(OptimalCutpoints)


We use the data frame called aSAH in pROC library :

data(aSAH)
aSAH1 <- aSAH[,c(2,6)]
rocobj <- roc(aSAH1$outcome, aSAH1$s100b, ci=T)


aSAH format :
A data frame containing 113 observations with 2 variables we will use :

outcome MCC type
s100b Concentration of S100B 




#Head of dataset
knitr::kable(head(aSAH1,8), align = "l", row.names = FALSE)
outcome s100b
Good 0.13
Good 0.14
Good 0.10
Good 0.04
Poor 0.13
Poor 0.10
Good 0.47
Poor 0.16

ggroc function

ggroc(
data, a roc object from the roc function, or a list of roc objects
aes, the name of the aesthetics for geom_line to map to the different ROC curves supplied. Use “group” if you want the curves to appear with the same aestetic, for instance if you are faceting instead
legacy.axes, a logical indicating if the specificity axis (x axis) must be plotted as as decreasing “specificity” (FALSE, the default) or increasing “1 - specificity” (TRUE) as in most legacy software
... )

Basic ROC curve

# roc basic
ggroc(rocobj) + 
  theme_minimal() #ggplot2 theme

Customized ROC curve

# add options
# confidence interval of the sensitivity
ci.se.rocobj <- ci(rocobj, of="se", boot.n=2000) 
dat.ci <- data.frame(x = as.numeric(rownames(ci.se.rocobj)),
                     lower = ci.se.rocobj[, 1],
                     upper = ci.se.rocobj[, 3])

# selection of a threshold
dat.seuil=coords(rocobj, "best", best.method=c("youden"), ret=c("threshold", "sensitivity", "specificity"))


# roc with options: CI, selected threshold, title with AUC and confidence interval
ggroc(rocobj) + 
  theme_minimal() +   #ggplot theme
  geom_abline(slope=1, intercept = 1, linetype = "dashed", alpha=0.7, color = "grey") +   # add abline and custom
  coord_equal() +   #  ensures the units are equally scaled on the x-axis and on the y-axis
  geom_ribbon(data = dat.ci, aes(x = x, ymin = lower, ymax = upper), fill = "grey", alpha= 0.2) +  # generate confidence interval of sensibility
  ggtitle(paste0("AUC=", round(rocobj$auc, 2), ", ", capture.output(rocobj$ci))) +   # plot title
  geom_point(data = tibble(se=dat.seuil$sensitivity, sp=dat.seuil$specificity), mapping = aes(x=sp, y=se), colour = "red") +
  annotate("text", x=dat.seuil$specificity-0.16, y=dat.seuil$sensitivity-0.05, 
           label= paste0(round(dat.seuil$threshold, 2), " (", round(dat.seuil$specificity, 2), ",", round(dat.seuil$sensitivity, 2), ")"))   # add anotations




Contact

This document is a work by Emma Lafaurie (emma.lafaurie@inserm.fr) for the SBIM (Service de Biostatistique et Information Médicale) at Saint-Louis Hospital in Paris.
Based on the template of Yan Holtz.