bencalibr: Plotting calibration for benefit of a prediction model

In predieval: Assessing Performance of Prediction Models for Predicting Patient-Level Treatment Benefit

Plotting calibration for benefit of a prediction model

Description

This function produces a plot to illustrate the calibration for benefit for a prediction model. The samples are split into a number of groups according to their predicted benefit, and within each group the function estimates the observed treatment benefit and compares it with the predicted one

Usage

bencalibr(
  data = NULL,
  Ngroups = 5,
  y.observed,
  treat,
  predicted.treat.0,
  predicted.treat.1,
  type = "continuous",
  smoothing.function = "lm",
  axis.limits = NULL
)

Arguments

data	An optional data frame containing the required information.
Ngroups	The number of groups to split the data.
y.observed	The observed outcome.
treat	A vector with the treatment assignment. This must be 0 (for control treatment) or 1 (for active treatment).
predicted.treat.0	A vector with the model predictions for each patient, under the control treatment. For the case of a binary outcome this should be probabilities of an event.
predicted.treat.1	A vector with the model predictions for each patient, under the active treatment. For the case of a binary outcome this should be probabilities of an event.
type	The type of the outcome, "binary" or "continuous".
smoothing.function	The method used to smooth the calibration line. Can be "lm", "glm", "gam", "loess", "rlm". More details can be found in https://ggplot2.tidyverse.org/reference/geom_smooth.html.
axis.limits	Sets the limits of the graph. It can be a vector of two values, i.e. the lower and upper limits for x and y axis. It can be omitted.

Value

The calibration plot

Examples

# continuous outcome
dat1=simcont(200)$dat
head(dat1)
lm1=lm(y.observed~(x1+x2+x3)*t, data=dat1)
dat.t0=dat1; dat.t0$t=0
dat.t1=dat1; dat.t1$t=1
dat1$predict.treat.1=predict(lm1, newdata = dat.t1) # predictions in treatment
dat1$predict.treat.0=predict(lm1, newdata = dat.t0) # predicions in control
bencalibr(data=dat1, Ngroups=10, y.observed, predicted.treat.1=predict.treat.1,
          predicted.treat.0=predict.treat.0, type="continuous", treat=t,
          smoothing.function = "lm", axis.limits = c(-1, 1.3))
# binary outcome
dat2=simbinary(500)$dat
head(dat2)
glm1=glm(y.observed~(x1+x2+x3)*t, data=dat2, family = binomial(link = "logit"))
dat2.t0=dat2; dat2.t0$t=0
dat2.t1=dat2; dat2.t1$t=1
dat2$predict.treat.1=predict(glm1, newdata = dat2.t1) # predictions in treatment
dat2$predict.treat.0=predict(glm1, newdata = dat2.t0) # predicions in control
bencalibr(data=dat2, Ngroups=6, y.observed, predicted.treat.1=expit(predict.treat.1),
          predicted.treat.0=expit(predict.treat.0), type="binary", treat=t,
          smoothing.function = "lm")

predieval documentation built on April 19, 2022, 5:08 p.m.