ROC curve

R code of the ggplot2 example

library(ggplot2)
library(dplyr)
library(pROC)

## Heart disease is any narrowing of a coronary artery (`num` above 0).
patients <- datarium::heartdisease |>
  filter(!is.na(thalach), !is.na(oldpeak)) |>
  mutate(disease = num > 0)

curve_of <- function(predictor, name) {
  r <- roc(patients$disease, patients[[predictor]], quiet = TRUE)
  tibble(test = sprintf("%s, AUC %.2f", name, auc(r)),
         fpr = 1 - r$specificities,
         tpr = r$sensitivities) |>
    arrange(fpr, tpr)
}
curves <- bind_rows(curve_of("oldpeak", "ST depression"),
                    curve_of("thalach", "Maximum heart rate"))

tests <- tibble(test = unique(curves$test), fpr = 0.45, tpr = c(0.24, 0.14))

p <- ggplot(curves, aes(x = fpr, y = tpr, colour = test)) +
  geom_abline(colour = "#7d7d7d", linewidth = 0.4, linetype = "dashed") +
  annotate("text", x = 0.62, y = 0.56, label = "Chance", hjust = 0, size = 3, colour = "#7d7d7d") +
  geom_path(linewidth = 0.7) +
  geom_text(data = tests, aes(label = test), hjust = 0, size = 3.2, fontface = "bold") +
  scale_x_continuous(labels = lex_number, expand = expansion(mult = c(0, 0.04))) +
  scale_y_continuous(labels = lex_number, expand = c(0, 0)) +
  labs(title = "Two measures detect heart disease about equally well",
       subtitle = "857 patients; each step of a curve is one threshold for a positive test",
       x = "False positive rate", y = "Sensitivity",
       caption = "Data: Heart Disease (Janosi et al. 1989), via the datarium R package") +
  theme_lex_viz(grid = "xy", arrows = "xy") +
  theme(legend.position = "none")