predict.mtlr {MTLR}

Predictions for MTLR

Description

Compute survival curves and other fitted values for a model generated by mtlr.

Usage

## S3 method for class 'mtlr'
predict(object, newdata, type = c("survivalcurve",
  "prob_times", "prob_event", "mean_time", "median_time"), add_zero = T,
  times = c(), ...)

Arguments

object

an object of class mtlr, generated by the mtlr.

newdata

an optional new dataframe for which to perform predictions using MTLR. If left empty, predictions will be performed using the dataset used to generate the original mtlr object – note that any error calculation on these predictions will be optimistic since this will only be the resubstitution error and not be representative of error on a new test set.

type

the type of prediction desired. Options are the survival curve for the time points selected by mtlr ("survivalcurve"), the survival curve for given times ("prob_times"), the probability of survival at the observations event time ("prob_event"), the mean survival time ("mean_time"), and the median survival time ("median_time").

For "survivalcurve" and "prob_times", the first column of the matrix returned will correspond to the time points and all other columns will be the observations survival probability at those associated time points. The index of a (row) observation in newdata will correspond to the ith + 1 column of the returned matrix.

If "prob_event" is chosen the response (event time) is required. For both "prob_event" and "prob_times", if the event time is larger than all of the time points used to build the mtlr model then the last (lowest) probability is used. For example, if the event time is 100 but the largest time point estimated by the mtlr model was 80 then the survival probability at 100 is equal to the survival probability at 80, i.e. S(100) = S(80).

For "mean_time", if survival curves do not extend to zero survival probability a linear extension is added (a linear line from (time = 0,probability = 1) to (time = ?, probability =0)). This is the same for "median_time" except the line need only extend to survival probability = 0.5. A mean/median survival time of Inf is returned for survival curves with all survival probabilities of 1.

add_zero

if TRUE, a time point of "0" and a survival probability of "1" will be added to all survival curves. Additionally, if add_zero is TRUE, type = "mean_time" will represent the average survival time overall but if FALSE, then "mean_time" will be reduced by roughly the value of the first time point. However, "median_time" and "prob_event" will be unchanged.

times

For prediction method "prob_times" you may specify the times at which to predict the survival probability for each row in newdata. This values defaults to all unique event times (both censored and uncensored) in the data on which the model was trained.

...

for future methods.

Value

The desired prediction type (a matrix or vector of predictions).

Note

The predictions generated by type = "survivalcurve" can be plotted using plotcurves – packages ggplot2 and reshape2 must be installed to use this function.

See Also

mtlr plotcurves

Examples

library(survival)
mod <- mtlr(Surv(time,status)~., data = lung)

#Here our predictions are on the data from which we trained so our results will be optimistic
# since they are produced from resubstitution as opposed to some new test set.
predict(mod, type = "survivalcurve")
predict(mod, type = "prob_event")
predict(mod, type = "median_time")
predict(mod, type = "mean_time")

#Notice the difference of about 59:
predict(mod, type = "mean_time", add_zero = FALSE)