riskRegression (version 2020.12.08)

autoplot.predictCox: Plot Predictions From a Cox Model

Description

Plot predictions from a Cox model.

Usage

# S3 method for predictCox
autoplot(
  object,
  type = NULL,
  ci = object$se,
  band = object$band,
  plot = TRUE,
  smooth = FALSE,
  digits = 2,
  alpha = NA,
  group.by = "row",
  reduce.data = FALSE,
  ylab = NULL,
  first.derivative = FALSE,
  ...
)

Arguments

object

Object obtained with the function predictCox.

type

[character] The type of predicted value to display. Choices are: "hazard" the hazard function, "cumhazard" the cumulative hazard function, or "survival" the survival function.

ci

[logical] If TRUE display the confidence intervals for the predictions.

band

[logical] If TRUE display the confidence bands for the predictions.

plot

[logical] Should the graphic be plotted.

smooth

[logical] Should a smooth version of the risk function be plotted instead of a simple function?

digits

[integer] Number of decimal places when displaying the values of the covariates in the caption.

alpha

[numeric, 0-1] Transparency of the confidence bands. Argument passed to ggplot2::geom_ribbon.

group.by

[character] The grouping factor used to color the prediction curves. Can be "row", "strata", or "covariates".

reduce.data

[logical] If TRUE only the covariates that does take indentical values for all observations are displayed.

ylab

[character] Label for the y axis.

first.derivative

[logical] If TRUE, display the first derivative over time of the risks/risk differences/risk ratios. (confidence intervals are obtained via simulation).

...

Additional parameters to cutomize the display.

Value

Invisible. A list containing:

• plot: the ggplot object.

• data: the data used to create the plot.

See Also

predictCox to compute cumulative hazard and survival based on a Cox model.

Examples

# NOT RUN {
library(survival)
library(ggplot2)

#### simulate data ####
set.seed(10)
d <- sampleData(1e2, outcome = "survival")
seqTau <- c(0,sort(unique(d$time[d$event==1])), max(d$time))

#### Cox model ####
m.cox <- coxph(Surv(time,event)~ X1 + X2 + X3,
                data = d, x = TRUE, y = TRUE)

## display baseline hazard
e.basehaz <- predictCox(m.cox)
autoplot(e.basehaz, type = "cumhazard")
autoplot(e.basehaz, type = "cumhazard", smooth = TRUE)
autoplot(e.basehaz, type = "cumhazard", smooth = TRUE, first.derivative = TRUE)

## display baseline hazard with type of event 
e.basehaz <- predictCox(m.cox, keep.newdata = TRUE)
autoplot(e.basehaz, type = "cumhazard")

## display predicted survival
pred.cox <- predictCox(m.cox, newdata = d[1:2,],
  times = seqTau, type = "survival", keep.newdata = TRUE)
autoplot(pred.cox)
autoplot(pred.cox, smooth = TRUE)
autoplot(pred.cox, group.by = "covariates")
autoplot(pred.cox, group.by = "covariates", reduce.data = TRUE)


## predictions with confidence interval/bands
pred.cox <- predictCox(m.cox, newdata = d[1:2,,drop=FALSE],
  times = seqTau, type = "survival", band = TRUE, se = TRUE, keep.newdata = TRUE)
res <- autoplot(pred.cox, ci = TRUE, band = TRUE, plot = FALSE)
res$plot + facet_wrap(~row)
res2 <- autoplot(pred.cox, ci = TRUE, band = TRUE, alpha = 0.1, plot = FALSE)
res2$plot + facet_wrap(~row)

#### Stratified Cox model ####
m.cox.strata <- coxph(Surv(time,event)~ strata(X1) + strata(X2) + X3 + X4,
                      data = d, x = TRUE, y = TRUE)

## baseline hazard
pred.baseline <- predictCox(m.cox.strata, keep.newdata = TRUE, type = "survival")
res <- autoplot(pred.baseline)
res$plot + facet_wrap(~strata, labeller = label_both)

## predictions
pred.cox.strata <- predictCox(m.cox.strata, newdata = d[1:5,,drop=FALSE],
                              time = seqTau, keep.newdata = TRUE, se = TRUE)

res2 <- autoplot(pred.cox.strata, type = "survival", group.by = "strata", plot = FALSE)
res2$plot + facet_wrap(~strata, labeller = label_both) + theme(legend.position="bottom")

## smooth version
autoplot(pred.cox.strata, type = "survival", group.by = "strata", smooth = TRUE, ci = FALSE)
# }