summary.prodlim {prodlim}

Summary method for prodlim objects.

Description

Summarizing the result of the product limit method in life-table format. Calculates the number of subjects at risk and counts events and censored observations at specified times or in specified time intervals.

Usage

## S3 method for class 'prodlim'
summary(
  object,
  times,
  newdata,
  max.tables = 20,
  surv = TRUE,
  cause,
  intervals = FALSE,
  percent = FALSE,
  format = "df",
  ...
)

Arguments

object

An object with class ‘prodlim’ derived with prodlim

times

Vector of times at which to return the estimated probabilities.

newdata

A data frame with the same variable names as those that appear on the right hand side of the 'prodlim' formula. Defaults to object$X.

max.tables

Integer. If newdata is not given the value of max.tables decides about the maximal number of tables to be shown. Defaults to 20.

surv

Logical. If FALSE report event probabilities instead of survival probabilities. Only available for object$model=="survival".

cause

For competing risk models. The event of interest for which predictions of the absolute risks are obtained by evaluating the cause-specific cumulative incidence functions at times.

intervals

Logical. If TRUE count events and censored in intervals between the values of times.

percent

Logical. If TRUE all estimated values are multiplied by 100 and thus interpretable on a percent scale.

format

Control format of output. Since May 2021, the result is a data.table and data.frame with attributes. When there are multiple covariate strata or competing risks, these are indicated by columns. Set format to "list" to get the old behaviour.

...

Further arguments that are passed to the print function.

Details

For cluster-correlated data the number of clusters at-risk are are also given. Confidence intervals are displayed when they are part of the fitted object.

Value

A data.frame with the relevant information.

Author(s)

Thomas A. Gerds [e-mail]

See Also

prodlim, summary.Hist

Examples

library(lava)
set.seed(17)
m <- survModel()
distribution(m,~age) <- uniform.lvm(30,80)
distribution(m,~sex) <- binomial.lvm()
m <- categorical(m,~z,K=3)
regression(m,eventtime~age) <- 0.01
regression(m,eventtime~sex) <- -0.4
d <- sim(m,50)
d$sex <- factor(d$sex,levels=c(0,1),labels=c("female","male"))
d$Z <- factor(d$z,levels=c(1,0,2),labels=c("B","A","C"))

# Univariate Kaplan-Meier
# -----------------------------------------------------------------------------------------
fit0 <- prodlim(Hist(time,event)~1,data=d)
summary(fit0)

## show survival probabilities as percentage and
## count number of events within intervals of a
## given time-grid:
summary(fit0,times=c(1,5,10,12),percent=TRUE,intervals=TRUE)

## the result of summary has a print function
## which passes ... to print and print.listof
sx <- summary(fit0,times=c(1,5,10,12),percent=TRUE,intervals=TRUE)
print(sx,digits=3)

## show absolute risks, i.e., cumulative incidences (1-survival)
summary(fit0,times=c(1,5,10,12),surv=FALSE,percent=TRUE,intervals=TRUE)

# Stratified Kaplan-Meier
# -----------------------------------------------------------------------------------------

fit1 <- prodlim(Hist(time,event)~sex,data=d)
print(summary(fit1,times=c(1,5,10),intervals=TRUE,percent=TRUE),digits=3)
# old behaviour
print(summary(fit1,times=c(1,5,10),intervals=TRUE,percent=TRUE,format="list"),digits=3)

summary(fit1,times=c(1,5,10),intervals=TRUE,percent=TRUE)

fit2 <- prodlim(Hist(time,event)~Z,data=d)
print(summary(fit2,times=c(1,5,10),intervals=TRUE,percent=TRUE),digits=3)

## Continuous strata (Beran estimator)
# -----------------------------------------------------------------------------------------
fit3 <- prodlim(Hist(time,event)~age,data=d)
print(summary(fit3,
              times=c(1,5,10),
              newdata=data.frame(age=c(20,50,70)),
              intervals=TRUE,
              percent=TRUE),digits=3)

## stratified Beran estimator
# -----------------------------------------------------------------------------------------
fit4 <- prodlim(Hist(time,event)~age+sex,data=d)
print(summary(fit4,
              times=c(1,5,10),
              newdata=data.frame(age=c(20,50,70),sex=c("female","male","male")),
              intervals=TRUE,
              percent=TRUE),digits=3)

print(summary(fit4,
              times=c(1,5,10),
              newdata=data.frame(age=c(20,50,70),sex=c("female","male","male")),
              intervals=TRUE,
              percent=TRUE),digits=3)

## assess results from summary
x <- summary(fit4,times=10,newdata=expand.grid(age=c(60,40,50),sex=c("male","female")))
cbind(names(x$table),do.call("rbind",lapply(x$table,round,2)))

x <- summary(fit4,times=10,newdata=expand.grid(age=c(60,40,50),sex=c("male","female")))

## Competing risks: Aalen-Johansen
# -----------------------------------------------------------------------------------------
d <- SimCompRisk(30)
crfit <- prodlim(Hist(time,event)~X1,data=d)
summary(crfit,times=c(1,2,5))
summary(crfit,times=c(1,2,5),cause=1,intervals=TRUE)
summary(crfit,times=c(1,2,5),cause=1)
summary(crfit,times=c(1,2,5),cause=1:2)


# extract the actual tables from the summary 
sumfit <- summary(crfit,times=c(1,2,5),print=FALSE)
sumfit$table[[1]] # cause 1
sumfit$table[[2]] # cause 2


# '