Convert from a 'timeline' data set format for survival data to the counting process form, and vice versa.
totimeline(formula, data, id, istate)
fromtimeline(formula, data, id, istate="istate")a data set of the proper form
a model formula with a Surv object on the left
data set in which to evaluate the formula
the name of the identifier variable, which will be searched
first in the data. Multiple rows for the same subject will
share the same id value.
for totimeline the name of the variable in the
counting process data set that contains the initial state. For
fromtimeline the variable name to use for the initial state.
Counting process style data sets are heavily used in the survival
package for both time-dependent covariates and multistate data. Each
row of the data will contain a time interval (t1, t2), status or state
at the end of the interval, covariate values that apply over the
interval, and an id variable.
A timeline data set will have a single time covariate, an id
variable, along with other covariate and outcome values that were
observed at that time point. If some covariates are observed at a
particular time point but others were not, these other values would be
missing for that row. (The exception are covariates that are constant,
like birthdate or a genetic marker, which will normally appear across
all rows).
A disadvantage of the counting process form is that it requires special
tools for manipulation, e.g., tmerge; timeline data sets are much
simpler in structure and thus can benefit from a much wider variety of
tools in their creation. They are also more direct wrt ensuring
validity: each row should encode what was actually observed at
that time point. Another potential advantage is for variables such as
diabetes, which might be used as an outcome in one model and a predictor
in another. This requires two separate variables in a counting process
data set, since covariates change at the beginning of a time interval
and outcomes happen at the end of it.
The conversion from timeline to counting process form uses the same
rules with respect to missing values as tmerge, it is in fact
what is used behind the scenes to do the conversion.
tmerge, survSplit