This package includes functions for parametrically estimating the mean number of events in the presence of competing events. Researchers often disregard subsequent events in the competing risk setting although the event of interest might be of recurrent nature, e.g. heart attacks, strokes, complications etc.. One estimate of interest in this situation is the mean number of events in the presence of competing events, which can be estimated using the functions provided in this package.
The estimation of the mean number of events requires two steps:
- The user need to fit a joint flexible parametric model using
JointFPM(). - The fitted model object needs to be passed to
predict()in order to estimate the mean number of events.
Please note that this package is currently under development and might change throughout the process.
Installation
For installing the package from CRAN please use
install.packages("JointFPM")If you would like to use the latest development version from GitHub please use
remotes::install_github("entjos/JointFPM")Short Example
We will use a dataset of bladder cancer recurrences for the following example. The dataset is included in the survival package and includes information on bladder cancer patients receiving three different treatments: placebo, Pyridoxine, and Thiotepa (cf. help(survival::bladder1)).In order to fit a joint FPM we first need to reshape the dataset into a stacked format, i.e. each observation needs to have one row for the competing event and possible multiple rows for the recurrent event. In the example below we use the data.table package for the data preparation.
# Load packages
library(JointFPM)
library(data.table) # For data preparations
# Load bladder cancer dataset from survival package
bldr_df <- as.data.table(survival::bladder1)
bldr_df <- bldr_df[, .(id, treatment, start, stop, status)]
# Define dataset for competing event times
bldr_ce <- bldr_df[, .SD[stop == max(stop)],
by = id]
bldr_ce[, `:=`(ce = 1,
re = 0,
event = as.numeric(status %in% 2:3),
start = 0)]
# Define dataset for bladder cancer recurrences
bldr_re <- bldr_df[,
`:=`(ce = 0,
re = 1,
event = as.numeric(status == 1))]
# Combine datasets into one stacked dataset
bldr_stacked <- rbindlist(list(bldr_ce, bldr_re))
bldr_stacked[, `:=`(pyridoxine = as.numeric(treatment == "pyridoxine"),
thiotepa = as.numeric(treatment == "thiotepa"))]
bldr_stacked$stop[bldr_stacked$stop == 0] <- 1 # Add one day survival
# Print stacked dataset
head(bldr_stacked) id treatment start stop status ce re event pyridoxine thiotepa
1: 1 placebo 0 1 3 1 0 1 0 0
2: 2 placebo 0 1 3 1 0 1 0 0
3: 3 placebo 0 4 0 1 0 0 0 0
4: 4 placebo 0 7 0 1 0 0 0 0
5: 5 placebo 0 10 3 1 0 1 0 0
6: 6 placebo 0 10 3 1 0 1 0 0The next step is to fit a joint flexible parametric model using the stacked dataset.
bldr_model <- JointFPM(Surv(time = start,
time2 = stop,
event = event,
type = 'counting') ~ 1,
re_model = ~ pyridoxine + thiotepa,
ce_model = ~ pyridoxine + thiotepa,
re_indicator = "re",
ce_indicator = "ce",
df_ce = 3,
df_re = 3,
tvc_ce_terms = list(pyridoxine = 2,
thiotepa = 2),
tvc_re_terms = list(pyridoxine = 2,
thiotepa = 2),
cluster = "id",
data = bldr_stacked)Based on the model we can predict the mean number of events at different time points and covariate patterns. Please note the estimation of confidence intervals for the mean number of events is computer intensive. The following code might take some minutes to run on your machine.
predict(bldr_model,
newdata = data.frame(pyridoxine = 1,
thiotepa = 0),
t = c(10, 20, 50))