survival (version 3.8-11)

lvcf: Last Value Carried Forward

Description

Replace missing values in a covariate with the last non-missing value, separately for each id.

Usage

lvcf(id, x, time, first= TRUE)

Value

an updated copy of x

Arguments

id

subject identifier

x

the covariate which will be modified

time

optional; used to sort observations within subject

first

if TRUE and x is 0/1 or logical, then if the first value for a subject is NA it is assumed to be 0 or FALSE, as appropriate

Author

Terry Therneau

Details

If an analysis includes a covariate that was measured repeatedly, cholesterol say, the data may often include visits where this was not measured. It is convenient in that case to use, for each subject, the most recently measured value. When a value is extended beyond the last measurement, rather that just replacement of intermediate values, there is the potential for bias, e.g., if the reason for discontinuation is predictive of the future trajectory of values.

Yes/no indicators of an event, e.g., onset of diabetes, are sometimes NA in a data set until the time point when they first occur. The first argument is in response to this.

If time is missing, then the order of the values within subject is assumed to be the calendar time order. It is not necessary for the the data to be sorted by id.

See Also

tmerge

Examples

newplat <- with(pbcseq, lvcf(id, platelet))
table(is.na(pbcseq$platelet), is.na(newplat))