Decision curve

Net benefit across decision thresholds. Net benefit weights the true positives a model finds against the false positives it costs, at the threshold probability a clinician would use to act. It is the figure that answers whether using the model is better than treating everyone or no one.

Usage

plot_decision_curve(
  probability,
  outcome,
  positive = NULL,
  thresholds = seq(0.01, 0.5, by = 0.01),
  treat_all = TRUE,
  treat_none = TRUE,
  palette = "house",
  xlab = "Threshold probability",
  ylab = "Net benefit",
  title = NULL,
  font = "Arial"
)

Arguments

probability

Predicted probabilities, or a matrix or data frame with one column per model.

outcome

Binary outcome, one value per observation.

positive

Value of outcome that represents the event.

thresholds

Decision thresholds to evaluate. Defaults to 0.01 to 0.5, which covers the range used in most clinical decisions.

treat_all

Draw the reference curve for treating every patient.

treat_none

Draw the reference line for treating no patient.

palette

Palette name passed to get_color().

xlab, ylab, title

Axis and plot labels.

font

Font family used in the plot.

Value

A ggplot object.

See also

plot_calibration() for whether the predicted levels are honest.

Examples

set.seed(1)
outcome <- rbinom(300, 1, 0.3)
probability <- plogis(rnorm(300) + outcome)
p <- plot_decision_curve(probability, outcome)