Create counting process data from timeline data
fromtimeline.RdThe primary survival functions accept any of simple, counting process, or timeline style arrangement of the data. This converts the third to the second
Usage
fromtimeline(formula, data, subset, id, repeated = FALSE, lvcf = TRUE,
yname = c("tstart", "tstop", "status"))Arguments
- formula
a formula with
Surv(time-variable, status-variable)as the response.- data
a data set currently in timeline form
- subset
optional expression to select a subset of rows
- id
the identifier variable, which rows go with each subject
- repeated
if the same level of the outcome repeats, without an intervening event of another type, should this be treated as a new event?
- lvcf
the last-value-carried-forward operation is applied to covariates on the right hand side
- yname
names to be used for the new response variable set, see details below
Details
The new data set will have fewer rows than the old, the row.names
can be used to map back to the original.
A new variable istate will be added, or _istate_ if
the former conflicts with a current variable.
If the yname vector is supplied then those names will be used
in the output. If not, and the response has a simple structure,
e.g. Surv(age, state) then the new data set will have variables
of age1, age2, state, and the response when modeling with
the new set would be Surv(age1, age2, state).
If the original response were more
complex, Surv(day, ifelse(day==0, 0, state2)) for instance,
then the third element of yname would be used for the status
variable.
If that that would generate a name conflict, i.e., the data set
already had a variable named "status", then "_status_" would be used as
the name.
The repeated argument controls how repeated instances of the same event
code are treated. If TRUE, they are treated as new events, an example
where this might be desired is repeated infections in a subject.
If FALSE, then repeats are not a new
event. An example would be a data set where we wanted to use
diabetes, say, as an endpoint, but this is diagnosis code is repeated
in the data at each medical visit.
The third option is repeated="first", in which case any repeated
instance of any level is treated as censored. In an data set with
states of cognitively unimpaired (CU), mild cognitive impairment
(MCI), dementia, and death, say, we might want an analysis with
"ever MCI" and "ever dementia" as the states. A subject with observed
states of (CU, CU, MCI, CU, MCI, MCI, dementia)
would have transitions of (CU, censor, MCI, CU, MCI, censor, dementia)
with repeated =FALSE and (CU, censor, MCI, censor, censor,
censor, death) with repeated = "first".
See the vignette on timeline data for a more complete discussion.